Economic Models - Convex Optimization
Economic Models - Convex Optimization
Economic Models - Convex Optimization
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
ECONOMIC MODELS<br />
Methods, Theory<br />
and Applications
This page intentionally left blank
ECONOMIC MODELS<br />
Methods, Theory<br />
and Applications<br />
editor<br />
Dipak Basu<br />
Nagasaki University, Japan<br />
World Scientific<br />
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • H O N G K O N G • TA I P E I • C H E N N A I
Published by<br />
World Scientific Publishing Co. Pte. Ltd.<br />
5 Toh Tuck Link, Singapore 596224<br />
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601<br />
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE<br />
British Library Cataloguing-in-Publication Data<br />
A catalogue record for this book is available from the British Library.<br />
ECONOMIC MODELS<br />
Methods, Theory and Applications<br />
Copyright © 2009 by World Scientific Publishing Co. Pte. Ltd.<br />
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,<br />
electronic or mechanical, including photocopying, recording or any information storage and retrieval<br />
system now known or to be invented, without written permission from the Publisher.<br />
For photocopying of material in this volume, please pay a copying fee through the Copyright<br />
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to<br />
photocopy is not required from the publisher.<br />
ISBN-13 978-981-283-645-8<br />
ISBN-10 981-283-645-4<br />
Typeset by Stallion Press<br />
Email: enquiries@stallionpress.com<br />
Printed in Singapore.
This Volume is Dedicated to the Memory of<br />
Prof. Tom Oskar Martin Kronsjo
This page intentionally left blank
Contents<br />
Tom Oskar Martin Kronsjo: A Profile<br />
About the Editor<br />
Contributors<br />
Introduction<br />
Chapter 1: Methods of Modelling: Mathematical<br />
Programming<br />
Topic 1: Some Unresolved Problems of Mathematical<br />
Programming<br />
by Olav Bjerkholt<br />
ix<br />
xiii<br />
xv<br />
xix<br />
1<br />
3<br />
Chapter 2: Methods of Modeling: Econometrics<br />
and Adaptive Control System<br />
21<br />
Topic 1: A Novel Method of Estimation Under Co-Integration 23<br />
by Alexis Lazaridis<br />
Topic 2: Time Varying Responses of Output to Monetary<br />
and Fiscal Policy<br />
by Dipak R. Basu and Alexis Lazaridis<br />
43<br />
Chapter 3: Mathematical Modeling in Macroeconomics 67<br />
Topic 1: The Advantages of Fiscal Leadership in an Economy<br />
with Independent Monetary Policies<br />
by Andrew Hughes Hallet<br />
Topic 2: Cheap-Talk Multiple Equilibria and Pareto —<br />
Improvement in an Environmental Taxation Games<br />
by Chirstophe Deissenberg and Pavel Ševčík<br />
69<br />
101<br />
vii
viii<br />
Contents<br />
Chapter 4: Modeling Business Organization 119<br />
Topic 1: Enterprise Modeling and Integration: Review<br />
and New Directions<br />
by Nikitas Spiros Koutsoukis<br />
Topic 2: Toward a Theory of Japanese Organizational Culture<br />
and Corporate Performance<br />
by Victoria Miroshnik<br />
121<br />
137<br />
Topic 3: Health Service Management Using Goal Programming 151<br />
by Anna-Maria Mouza<br />
Chapter 5: Modeling National Economies 171<br />
Topic 1: Inflation Control in Central and Eastern European<br />
Countries<br />
by Fabrizio Iacone and Renzo Orsi<br />
Topic 2: Credit and Income: Co-Integration Dynamics<br />
of the US Economy<br />
by Athanasios Athanasenas<br />
173<br />
201<br />
Index 223
Tom Oskar Martin Kronsjo: A Profile<br />
Lydia Kronsjo<br />
University of Birmingham, England<br />
Tom Kronsjo was born in Stockholm, Sweden, on 16 August 1932, the<br />
only child of Elvira and Erik Kronsjo. Both his parents were gymnasium<br />
teachers.<br />
In his school days Tom Kronsjo was a bright, popular student, often<br />
a leader among his school contemporaries and friends. During his school<br />
years, he organized and chaired a National Science Society that became<br />
very popular with his school colleagues.<br />
From an early age, Tom Kronsjo’s mother encouraged him to learn<br />
foreign languages. He put these studies to a good test during a 1948 summer<br />
vacation when he hitch-hiked through Denmark, Germany, the Netherlands,<br />
England, Scotland, Ireland, France, and Switzerland. The same year Tom<br />
Kronsjo founded and chaired a technical society in his school. In the summer<br />
vacation of 1949, Tom Kronsjo organized an expedition to North Africa<br />
by seven members of the technical society, their ages being 14–24. The<br />
young people traveled by coach through the austere post-Second World War<br />
Europe. All the equipment for travel, including the coach, were donated<br />
by sympathetic individuals and companies, who were impressed by an<br />
intelligent, enthusiastic group of youngsters, eager to learn about the world.<br />
This event caught a wide interest in the Swedish press at that time and was<br />
vividly reported.<br />
Tom Kronsjo used to mention that as a particularly personally exciting<br />
experience of his school years he remembered volunteering to give a series<br />
of morning school sermons on the need for the world governmental organizations,<br />
which would encourage and support mutual understanding of the<br />
needs of the people to live in peace.<br />
Tom Kronsjo’s university years were spent at both sides of the great<br />
divide. Preparing himself for a career in international economics, he wanted<br />
to experience and see for himself the different nations’ ways of life and<br />
points of view. He studied in London, Belgrade,Warsaw, Oslo, and Moscow,<br />
before returning to his home country, Sweden, to pass examinations and<br />
ix
x<br />
Tom Oskar Martin Kronsjo<br />
submit required work at the Universities of Uppsala, Stockholm and Lund<br />
for his Fil.kand. degree in economics, statistics, mathematics and Slavonic<br />
Languages, and Fil.lic. degree in economics. By then, Tom Kronsjo besides<br />
his mother tongue — Swedish, was fluent in six languages — German,<br />
English, French, Polish, Serbo-Croatian, and Russian. During this period,<br />
Tom was awarded a number of scholarships for study in Sweden and abroad.<br />
Tom Kronsjo met his wife while studying in Moscow. He was a visiting<br />
graduate student at the Moscow State University and she a graduate student<br />
with the Academy of Sciences of the USSR. They married in Moscow in<br />
February 1962. A few years later, a very happy event in the family life was<br />
the arrival of their adoptive son from Korea in October 1973.<br />
As a post-graduate student in economics, Tom Kronsjo was one of a<br />
new breed of Western social scientists that applied rigorous mathematical<br />
and computer-based modeling to economics. In 1963, Tom Kronsjo was<br />
invited by Professor Ragnar Frish, later Nobel prize winner in econometrics,<br />
to work under him and Dr. Salah Hamid in Egypt as a visiting Assistant<br />
Professor in Operations Research at the UAR Institute of National Planning.<br />
His pioneering papers on optimization of foreign trade were soon<br />
considered classic and brought invitations to work as a guest researcher by<br />
the GDR Ministry of Foreign Trade and Inner German Trade, and by the<br />
Polish Ministry of Foreign Trade.<br />
In 1964, Tom Kronsjo was appointed an Associate Professor in Econometrics<br />
and Social statistics at the Faculty of Commerce and Social Sciences,<br />
the University of Birmingham, U.K. He and his wife moved to the<br />
United Kingdom.<br />
In Birmingham, together with Professor R.W. Davies Tom Kronsjo<br />
developed diploma, and master of social sciences courses in national economic<br />
planning. The courses, offered from 1967, attracted many talented<br />
students from all over the world. Many of these students then stayed<br />
with Tom Kronsjo as research students and many of these people are<br />
today university professors, industrialists, and economic advisors. In 1969,<br />
at the age of 36, Tom Kronsjo was appointed Professor of <strong>Economic</strong><br />
Planning. By then Tom Kronsjo had published over a hundred research<br />
papers.<br />
Tom Kronsjo has always had a strong interest in educational innovations.<br />
He was among the first professors at the university to introduce a<br />
television-based teaching, making it possible to widen the scope of material<br />
taught.<br />
Tom Kronsjo was generous towards his students with his research ideas,<br />
stimulating his students intellectually, unlocking wide opportunities for his
Tom Oskar Martin Kronsjo<br />
students through this generosity. He had the ability to see the possibilities<br />
in any situation and taught his students to use them.<br />
Tom Kronsjo’s research interests were wide and multifaceted. One<br />
area of his interest was planning and management at various levels of<br />
the national economy. He developed important methods of decomposition<br />
of large economic systems, which were considered by his contemporaries<br />
a major contribution to the theory of mathematical planning and<br />
programming. These results were published in 1972 in a book “<strong>Economic</strong>s<br />
of Foreign Trade” written jointly by Tom Kronsjo, Dr. Z. Zawada, and<br />
Professor J. Krynicki, both of the Warsaw University, and published in<br />
Poland.<br />
In 1972, Tom Kronsjo was invited as a Visiting Professor of <strong>Economic</strong>s<br />
and Industrial Administration at Purdue University, Purdue, USA.<br />
In 1974, Tom Kronsjo was a Distinguished Visiting Professor of <strong>Economic</strong>s,<br />
at San Diego State University, San Diego, USA.<br />
In the year 1975, Tom Kronsjo found himself as a Senior Fellow in the<br />
Foreign Policy Studies Program of the Brookings Institution, Washington,<br />
D.C., USA. In collaboration with three other scientists, he was engaged<br />
in the project entitled “Trade and Employment Effects of the Multilateral<br />
Trade Negotiations”. In the words of his colleagues, Tom Kronsjo made an<br />
indispensable contribution to the project. Tom Kronsjo’s ingenious, tireless,<br />
conceptual, and implementing efforts made it possible to carry out<br />
calculations involving millions of pieces of information on trade and tariffs,<br />
permitting the Brookings model to become the most sophisticated and comprehensive<br />
model available at the time for investigation into effects of the<br />
current multilateral trade negotiations. In addition, Tom Kronsjo conceived<br />
and designed the means for implementing, perhaps the most imaginative<br />
portion of the project, the calculation of “optimal” tariff cutting formulas<br />
using linear programming and taking account of balance of payments and<br />
other constraints on trade liberalization. A monograph on “Trade, Welfare,<br />
and Employment Effects of Trade Negotiations in the Tokyo Round” by<br />
Dr. W.R. Cline, Dr. Noboru Kawanabe, Tom Kronsjo, and T. Williams was<br />
published in 1978.<br />
Tom Kronsjo’s general interests have been in the field of conflicts and<br />
co-operation, war and peace, ways out of the arms race, East-West joint<br />
enterprises and world development. He published papers on unemployment,<br />
work incentives, and nuclear disarmament.<br />
Tom Kronsjo served as one of the five examiners for one of the Nobel<br />
prizes in economic sciences. From time to time, he was invited to nominate<br />
a candidate or candidates for the Nobel prize in economics.<br />
xi
xii<br />
Tom Oskar Martin Kronsjo<br />
Tom Kronsjo took early retirement from the University of Birmingham<br />
in 1988 and devoted subsequent years working with young researchers from<br />
China, Russia, Poland, Estonia, and other countries.<br />
Tom Kronsjo was a great enthusiast of vegetarianism and veganism,<br />
of Buddhist philosophy, yoga, and inter-space communication, of aviation<br />
and flying. He learned to fly in the United States and held a pilot license<br />
since 1972.<br />
Tom Kronsjo was a man who lived his life to the fullest, who constantly<br />
looked for ways to bring peace to the world. He was an exceptional individual<br />
who had the incredible vision and an ability to foresee new trends.<br />
He was a man of great charisma, one of those people of whom we say “he<br />
is a man larger than life”.<br />
Tom Oskar Martin Kronsjo died in June 2005 at the age of 72.
About the Editor<br />
Prof. Dipak Basu is currently a Professor in International <strong>Economic</strong>s in<br />
Nagasaki University in Japan. He has obtained his Ph.D. from the University<br />
of Birmingham, England and did his post-doctoral research in Cambridge<br />
University. He was previously a Lecturer in Oxford University — Institute<br />
of Agricultural <strong>Economic</strong>s, Senior Economist in the Ministry of Foreign<br />
Affairs, Saudi Arabia and Senior Economist in charge of Middle East &<br />
Africa Division of Standard & Poor (Data Resources Inc). He has published<br />
six books and more than 60 research papers in leading academic journals.<br />
xiii
This page intentionally left blank
Contributors<br />
Athanasios Athanasenas is an Assistant Professor of Operations Research<br />
& <strong>Economic</strong> Development in the School of Administration & <strong>Economic</strong>s,<br />
Institute of Technology and Education, Serres, Greece. He was educated in<br />
Colorado and Virginia Universities and received his Ph.D. from the University<br />
of Minnesota. He was a Full-Bright scholar.<br />
Olav Bjerkholt, Professor in <strong>Economic</strong>s, University of Oslo, was previously<br />
the Head of the Research at the Norwegian Statistical Office and an<br />
Assistant Professor of Energy <strong>Economic</strong>s at the University of Oslo. He has<br />
obtained his Ph.D. from the University of Oslo. He was a Visiting Professor<br />
in the Massachusetts Institute of Technology and was a consultant to the<br />
United Nations and the IMF.<br />
Christophe Deissenberg is a Professor in <strong>Economic</strong>s, at Université dela<br />
Méditerranée and GREQAM, France. He is a co-editor of the Journal of<br />
<strong>Optimization</strong> Theory and Applications and of Macroeconomic Dynamics,<br />
a member of the editorial board of Computational <strong>Economic</strong>s and of Computational<br />
Management Science. He is a member of the Advisory Board of<br />
the Society for Computational <strong>Economic</strong>s and a member of the Management<br />
Committee of the ESF action COSTP10–Physics of Risk, and a team<br />
leader of the European STREP project EURACE.<br />
Andrew Hughes Hallett is a Professor of <strong>Economic</strong>s and Public Policy<br />
in the School of Public Policy at George Mason University, USA. Previously<br />
he was a Professor of <strong>Economic</strong>s at Vanderbilt University, and the<br />
University of Strathclyde in Glasgow, Scotland. He was educated in the<br />
University of Warwick and London School of <strong>Economic</strong>s, holds a doctorate<br />
from Oxford University. He was a Visiting Professor in Princeton<br />
University, Free University of Berlin, Universities of Warwick, Frankfurt,<br />
Rome, Paris X and in the Copenhagen Business School. He is a Fellow of<br />
the Royal Society of Edinburgh.<br />
xv
xvi<br />
Contributors<br />
Fabrizio Iacone obtained a Ph.D. in <strong>Economic</strong>s at the London School of<br />
<strong>Economic</strong>s and is currently a Lecturer in the Department of <strong>Economic</strong>s at<br />
the University of York, UK. He has already published several articles in<br />
leading journals.<br />
Nikos Konidaris is currently a Ph.D. student at Hellenic Open University,<br />
School of Social Sciences, researching in the field of Business Administration.<br />
He holds an MA degree in Business and Management and a Bachelor’s<br />
Degree in <strong>Economic</strong>s & Marketing from Luton University UK.<br />
Nikitas Spiros Koutsoukis is anAssistant Professor of Decision Modelling<br />
and Scientific Management in the Department of International <strong>Economic</strong><br />
Relations and Development, Democritus University, Greece. He holds MSc<br />
and Ph.D. in decision modeling from Brunel University (UK). He is interested<br />
mainly in decision modeling and governance systems, with applications<br />
in risk management, business intelligence and operational research<br />
based decision support.<br />
Alexis Lazaridis, Professor in Econometrics, Aristotle University of Thessaloniki,<br />
Greece, has obtained his Ph.D. from the University of Birmingham,<br />
England. He was a member of the Administrative Board of the Organization<br />
of Mediation and Arbitration, of the Administrative Board of the<br />
Centre of International and European <strong>Economic</strong> Law, and of the Supreme<br />
Labour Council of Greece. During his term of service as Head of the <strong>Economic</strong><br />
Department Faculty of Law and <strong>Economic</strong>s, Aristotle University of<br />
Thessaloniki, he was conferred an honorary doctoral degree by the President<br />
of the Republic of Cyprus. He was also pronounced an honorary<br />
member of the Centre of International and <strong>Economic</strong> European Law by the<br />
Secretary-General of the Legal Services of E.U. He has published several<br />
books, large number of scientific papers and has patents on two computer<br />
software programs.<br />
Athanassios Mihiotis is an Assistant Professor at the School of Social<br />
Science of the Hellenic Open University. He has obtained his Ph.D. in<br />
industrial management from the National Technical University of Athens,<br />
Greece. He is the editor of the International Journal of Decision Sciences,<br />
Risk and Management. He has in the past worked as Planning and Logistics<br />
Director for multinational and domestic companies and has served as<br />
member of project teams in the Greek public sector.
Contributors<br />
Victoria Miroshnik, is currently anAdam Smith Research Scholar, Faculty<br />
of Law and Social Sciences, University of Glasgow, Scotland. She was<br />
educated in Moscow State University and was an Assistant Professor in<br />
Tbilisi State University, Georgia. She has published two books and a large<br />
number of scientific papers in leading management journals.<br />
Anna-Maria Mouza, Assistant Professor in Business Administration,<br />
Technological Educational Institute of Serres, Greece, has obtained her<br />
Ph.D. from the School of Health Sciences, Aristotle University of<br />
Thessaloniki, Greece. Previously, she was an Assistant Professor in the<br />
School of Technology Applications of the Technological Institute of<br />
Thessaloniki, in the Department of Agriculture and in the Faculty of Health<br />
Sciences of the Aristotle University of Thessaloniki. She is the author of<br />
two books and many published scientific papers.<br />
R. Orsi, Professor in Econometrics, University of Bologna, Italy, has<br />
obtained his Ph.D. from Universitè Catholique de Louvain, Belgium. He<br />
was previously educated in the University of Bologna and in the University<br />
of Rome. He was an Associate Professor in the University of Modena, and<br />
Professor of Econometrics in the University of Calabria, Italy. He was a<br />
Visiting Professor in the Center for Operation Research and Econometrics<br />
(CORE), Universitè-Catholique de Louvain, Belgium, Nato Senior Fellow,<br />
Head of the Department of <strong>Economic</strong>s at the University of Calabria<br />
from 1987 to 1989, Director of CIDE (the Inter-Universitary Center of<br />
Econometrics) from 1997 to1999, and from 2000 to 2002, and the Head of<br />
Department of <strong>Economic</strong>s, University of Bologna from 1999 to 2002.<br />
Pavel Ševčík, was educated in the Université delaMéditerranée. He is<br />
presently completing his Ph.D in the Universilé de Montréal, Canada.<br />
xvii
Introduction<br />
We define “model building in economics,” as the fruitful area of economics<br />
designed to solve real-world problems using all available methods available<br />
without distinctions: mathematical, computational, and analytical.<br />
Wherever needed we should not be shy to develop new techniques —<br />
whether mathematical or computational. This was the philosophy of Prof<br />
Tom Kronsjo, in whose memory we dedicate this volume. The idea is to<br />
develop stylized facts amenable to analytical methods to solve problems of<br />
the world.<br />
The articles in this volume are divided into three distinct groups: methods,<br />
theory, and applications.<br />
In the method section there are two parts: mathematical programming<br />
and computation of co-integrating vectors, which are widely used in econometric<br />
analysis. Both these parts are important to analyze and develop policies<br />
in any analytical model. Berjkholt has reviewed the history of development<br />
of mathematical programming and its impacts in economic modeling.<br />
In the process, he has drawn our attention to some methods, first developed<br />
by Ragner Frisch, which can provide solution where the standard Simplex<br />
method may fail. Frisch has developed these methods while working on the<br />
Second Five Year’s Plan for India and has developed a method, which is<br />
similar to that of Karmarkar.<br />
Lazaridis presents a completely new method to compute co-integration<br />
vectors, widely used in econometric analysis, by applying singular value<br />
decomposition. With this method, one can easily accommodate in the cointegrating<br />
vectors any deterministic factors, such as dummies, apart from<br />
the constant term and the trend. In addition, a comparatively simple procedure<br />
is developed for determining the order of integration of a different stationary<br />
series. Besides, with this procedure one can directly detect whether<br />
the differencing process produces a stationary series or not, since it seems<br />
to be a common belief that differencing a variable (one or more times) we<br />
will always get a stationary series, although this is not necessarily the case.<br />
Basu and Lazaridis have proposed a new method of estimation and<br />
solution of an econometric model with parameters moving over time. This<br />
xix
xx<br />
Introduction<br />
type of model is very realistic for economies, which are in transitional<br />
phases. The model was applied for the Indian economy to derive monetaryfiscal<br />
policy structure that would respond to the changing environments of<br />
the economy. The authors provide analysis to show that the response of the<br />
economy would be variable i.e., not at all fixed over time.<br />
In the macroeconomic theory section Hughes-Hallet has examined the<br />
impacts of fiscal policy in a regime with independent monetary authority<br />
and how to co-ordinate public policies in such a regime. Independence<br />
of the central bank is a crucial issue for both developed and developing<br />
country. In this theoretical model, the author has discovered some important<br />
conclusions that fiscal leadership leads to improved outcomes because it<br />
implies a degree of co-ordination and reduced conflicts between institutions,<br />
without the central bank having to lose its ability to act independently.<br />
Deissenberg and Pavel consider a dynamic model of environmental<br />
taxation that exhibits time inconsistency. There are two categories of firms,<br />
believers (who take the non-binding tax announcements made by the regulator<br />
at face value), and non-believers (who make rational but costly predictions<br />
of the true regulator’s decisions). The proportion of believers and<br />
non-believers changes over time depending on the relative profits of both<br />
groups. If the regulator is sufficiently patient and if the firms react sufficiently<br />
fast to the profit differences, multiple equilibria can arise. Depending<br />
on the initial number of believers, the regulator will use cheap-talk together<br />
with actual taxes to steer the economy either to the standard, perfect prediction<br />
solution suggested in the literature, or to equilibrium with a mixed<br />
population of believers and non-believers who both make prediction errors.<br />
This model can be very useful in the analysis of environmental economics.<br />
In the section on theory of business organization, Victoria Miroshnik<br />
has formulated a model of Japanese organization and how the superior<br />
corporate performances in the Japanese multi-national companies are the<br />
result of a multi-layer structure of cultures, national, personal, organizational.<br />
The model goes deep into the analysis of fundamental ingredient of<br />
Japanese organizational culture and human resource management system<br />
and how these contribute to corporate performance. The model produces a<br />
novel scheme, how to estimate phenomena which are seemingly unobserved<br />
psychological and sociological characteristics.<br />
Mihiotis et al., discussed enterprise modeling and integration challenges,<br />
and rationale as well as current tools and methodology for deeper<br />
analysis. Over the last decades, much has been written and a lot of research<br />
has been undertaken on the issue of enterprise modeling and integration.<br />
The purpose of this discussion is to provide an overview and understanding
Introduction<br />
of the key concepts. The authors have outlined a framework for advancing<br />
enterprise integration modeling based on the state-of-the art techniques.<br />
Anna Maria Mauza in a path breaking research presents a model suitable<br />
for an efficient budget management of a health service unit, by applying goal<br />
programming. She analyzes all the details needed to formulate the proper<br />
model, in order to successfully apply goal programming, in an attempt to<br />
satisfy the expectations of the decision maker in the best possible way,<br />
providing at the same time alternative scenarios considering various socioeconomic<br />
factors.<br />
In the section for policy analysis, Iacone and Orsi applied a small macroeconometric<br />
model to ascertain if the inflation dynamics and controls for<br />
Poland, Czech Republic, and Slovenia, are compatible with the remaining<br />
EU member countries. They found that the real exchange rate is the most<br />
effective instrument to stabilize inflation whereas direct inflation control<br />
mechanisms may be ineffective in certain cases. These experiments are<br />
very useful to design anti-inflation policies in open economies.<br />
AthanasiosAthanasenas investigated the co-integration dynamics of the<br />
credit–income nexus, within the economic growth process of the post-war<br />
US economy, over the period from 1957 up to 2007. Given the existing<br />
empirical research on the credit-lending channel and the established relationship<br />
between financial intermediation and economic growth in general,<br />
the main purpose is to analyze in detail the causal relationship between<br />
finance and growth by focusing on bank credit and income GNP, in the<br />
post-war US economy. This is a new application of an innovative technique<br />
of co-integration analysis with emphasis on system stability analysis. The<br />
results show that there is no short-run effect of credit changes on income<br />
changes, but only in the long-run, credit affects money income.<br />
The book covers most of the important areas of economics with the basic<br />
analytical framework to formulate a logical structure and then suggest and<br />
implement methods to quantify the structure to derive applicable policies.<br />
We hope the book would be a source of joy for anyone interested to make<br />
economics a useful discipline to enhance human welfare rather than being<br />
a sterile discourse devoid of reality.<br />
xxi
This page intentionally left blank
Chapter 1<br />
Methods of Modelling: Mathematical<br />
Programming
This page intentionally left blank
Topic 1<br />
Some Unresolved Problems of Mathematical<br />
Programming<br />
Olav Bjerkholt<br />
University of Oslo, Norway<br />
1. Introduction<br />
Linear programming (LP) emerged in the United States in the early postwar<br />
years. One may to a considerable degree see the development of LP<br />
as a direct result of the mobilization of research efforts during the war.<br />
George B. Dantzig, who was employed by the US Armed Forces, played a<br />
key role in developing the new tool by his discovery in 1947 of the Simplex<br />
method for solving LP problems. Linear programming was thus a military<br />
product, which soon appeared to have very widespread civilian applications.<br />
The US Armed Forces continued its support of Dantzig’s LP work, as the<br />
most widespread textbook in LP in the 1960s, namely Dantzig (1963), was<br />
sponsored by the US Air Force.<br />
Linear programming emerged at the same time as the first electronic<br />
computers were developed. The uses and improvements in LP as an optimization<br />
tool developed in step with the fast development of electronic<br />
computers.<br />
A standard formulation of a LP problem in n variables is as follows:<br />
min{c ′ x |Ax = b, x ≥ 0}<br />
x<br />
where c ∈ R n , b ∈ R m and A a (m, n) matrix of rank m
4 Olav Bjerkholt<br />
Among these was also Ragnar Frisch. Frisch was broadly oriented towards<br />
macroeconomic policy and planning problems and was highly interested<br />
in the promising new optimization tool. Frisch, who had a strong background<br />
in mathematics and also was very proficient in numerical methods,<br />
made the development of solution algorithms for LP problems a part of his<br />
research agenda. During the 1950s, Frisch invented and tested out various<br />
solution algorithms along other lines than the Simplex method for solving<br />
LP problems.<br />
With regard to the magnitude of LP problems that could be solved<br />
at different times with the computer equipment at disposal, Orden (1993)<br />
gives the following rough indication. In the first year after 1950, the number<br />
m of constraints in a solvable problem was of order 100 and has grown<br />
with a factor of 10 for each decade, implying that currently LP problems<br />
may have a number of constraints that runs into tens of millions. Linear<br />
programming has been steadily taken into use for new kinds of problems<br />
and the computer development has allowed the problems to be bigger and<br />
bigger.<br />
The Simplex method has throughout this period been the dominant<br />
algorithm for solving LP problems, not least in the textbooks. It is perhaps<br />
relatively rare that an algorithm developed at an early stage for new problem,<br />
as Dantzig did with the Simplex method for the LP problem, has retained<br />
such a position. The Simplex method will surely survive in textbook presentations<br />
and for its historical role, but for the solution of large-scale LP<br />
problems it is yielding ground to alternative algorithms. Ragnar Frisch’s<br />
work on algorithms is interesting in this perspective and has been given<br />
little attention. It may on a modest scale be a case of a pioneering effort<br />
that was more insightful than considered at the time and thus did not get<br />
the attention it deserved. In the history of science, there are surely many<br />
such cases.<br />
Let us first make a remark on the meaning of algorithms. The mathematical<br />
meaning of “algorithm” in relation to a given type of problem is a<br />
procedure, which after finite number of steps finds the solution to the problem,<br />
or determines that there is no solution. Such an algorithmic procedure<br />
can be executed by a computer program, an idea that goes back to Alan<br />
Turing. But when we talk about algorithms with regard to the LP problem<br />
it is not in the mathematical sense, but a more practical one. An LP<br />
algorithm is a practically executable technique for finding the solution to<br />
the problem. Dantzig’s Simplex method is an algorithm in both meanings.<br />
But one may have an algorithm in the mathematical sense, which is not a<br />
practical algorithm (and indeed also vice versa).
Some Unresolved Problems 5<br />
In the mathematical sense of algorithm, the LP problem is trivial. It<br />
can easily be shown that the feasibility region is a convex set with a linear<br />
surface. The optimum point of a linear criterion over such a set must be in a<br />
corner or possibly in a linear manifold of dimension greater than one.As the<br />
number of corners is finite, the solution can be found, for example, by setting<br />
n − m of the x’s equal to zero and solve Ax = b for the remaining ones.<br />
Then all the corners can be searched for the lowest value of the optimality<br />
criterion.<br />
Dantzig (1984) gives a beautiful illustration of why such a search for<br />
optimality is not viable. He takes a classical assignment problem, the distribution<br />
of a given number of jobs among the same number of workers.<br />
Assume that 70 persons shall be assigned to 70 jobs and the return of each<br />
worker-job combination is known. There are thus 70! possible assignment<br />
combinations. Dantzig’s comment about the possibility of looking at all<br />
these combinations runs as follows:<br />
“Now 70! is a big number, greater than 10 100 . Suppose we had an IBM<br />
370-168 available at the time of the big bang 15 billion years ago. Would<br />
it have been able to look at all the 70! combinations by the year 1981?<br />
No! Suppose instead it could examine 1 billion assignments per second?<br />
The answer is still no. Even if the earth were filled with such computers<br />
all working in parallel, the answer would still be no. If, however, there<br />
were 10 50 earths or 10 44 suns all filled with nano-second speed computers<br />
all programmed in parallel from the time of the big bang until sun grows<br />
cold, then perhaps the answer is yes.” (Dantzig, 1984, p. 106)<br />
In view of these enormous combinatorial possibilities, the Simplex<br />
method is a most impressive tool by making the just mentioned and even<br />
bigger problems practically solvable. The Simplex method is to search for<br />
the optimum on the surface of the feasible region, or, more precisely, in the<br />
corners of the feasible region, in such a way that the optimality criterion<br />
improves at each step.<br />
A completely different strategy to search for the optimum is to search<br />
the interior of the feasible region and approach the optimal corner (or one of<br />
them) from within, so to speak. It may seem to speak for the advantage of the<br />
Simplex method that it searches an area where the solution is known to be.<br />
A watershed in the history of LP took place in 1984 when Narendra<br />
Karmarkar’s algorithm was presented at a conference (Karmarkar, 1984a)<br />
and published later the same year in a slightly revised version (Karmarkar,<br />
1984b). Karmarkar’s algorithm, which the New York Times found worthy<br />
as first page news as a great scientific breakthrough, is such an “interior”
6 Olav Bjerkholt<br />
method, searching the interior of the feasibility area in the direction of the<br />
optimal point.<br />
This idea had, however, been pursued by Ragnar Frisch. To the best of<br />
my knowledge this was first noted by Roger Koenker a few years ago:<br />
“But it is an interesting irony, illustrating the spasmodic progress of science,<br />
that the most fruitful practical formulation of the interior point revolution<br />
of Karmarkar (1984) can be traced back to a series of Oslo working<br />
papers by Ragnar Frisch in the early 1950s.” (Koenker, 2000).<br />
2. Linear Programming in <strong>Economic</strong>s<br />
Linear programming has a somewhat curious relationship with academic<br />
economics. Few would today consider LP as a part of economic science. But<br />
LP was, so to say, launched within economics, or even within econometrics,<br />
and given much attention by leading economists for about 10 years or so.<br />
After around 1960, the ties to economics were severed. Linear programming<br />
disappeared from the economics curriculum and lost the attention of<br />
academic economists. It belonged from then on to operations research and<br />
management science on one hand and to computer science on the other.<br />
It is hardly possible to answer to give an exact date for when “LP”<br />
was born or first appeared. It originated at around the same time as game<br />
theory with which it shares some features, and also at the time of some<br />
applied problems such as the transportation problem and the diet problem.<br />
Linear programming has various roots and forerunners in economics and<br />
mathematics in attempts to deal with economic or other problems using<br />
linear mathematical techniques. One such forerunner, but only slightly, was<br />
Wassily Leontief’s input-output analysis. Leontief developed his “closed<br />
model” in the 1930s in an attempt to give empirical content to Walrasian<br />
general equilibrium. Leontief’s equilibrium approach was transformed in<br />
his cooperation with the US Bureau of Labor Statistics to the open inputoutput<br />
model, see Kohli (2001).<br />
In the early post-war years, LP and input-output analysis seemed within<br />
economics to be two sides of the same coin. The two terms were, for<br />
a short term, even used interchangeably. The origin of LP could be set<br />
to 1947, which, as already mentioned, was when Dantzig developed the<br />
Simplex algorithm. If we state the question as to when “LP” was first<br />
used in its current meaning in the title of a paper presented at a scientific<br />
meeting, the answer is to the best of my knowledge 1948. At<br />
the Econometric Society meeting in Cleveland at the end of December
Some Unresolved Problems 7<br />
1948, Leonid Hurwicz presented a paper on LP and the theory of optimal<br />
behavior with the first paragraph providing a concise definition for<br />
economists:<br />
“The term linear programming is used to denote a problem of a type<br />
familiar to economists: maximization (or minimization) under restrictions.<br />
What distinguishes linear programming from other problems in<br />
this class is that both the function to be maximized and the restrictions<br />
(equalities or inequalities) are linear in the variables.” (Hurwicz, 1949).<br />
Hurwicz’s paper appeared in a session on LP, which must have been<br />
the first ever with that title. One of the other two papers in the session was<br />
by Wood and Dantzig and discussed the problem of selecting the “best”<br />
method of accomplishing any given set of objectives within given restriction<br />
as a LP problem, using an airlift operation as an illustrative example. The<br />
general model was presented as an “elaboration of Leontief’s input-output<br />
model.”<br />
At an earlier meeting of the Econometric Society in 1948, Dantzig had<br />
presented the general idea of LP in a symposium on game theory. Koopmans<br />
had, at the same meeting, presented his general activity analysis production<br />
model. The conference papers of both Wood and Dantzig appeared in<br />
Econometrica in 1949. Dantzig (1949) mentioned as examples of problems<br />
for which the new technique could be used, Stigler (1945), known in the<br />
literature as having introduced the “diet problem,” and a paper by Koopmans<br />
on the “transportation problem,” presented at an Econometric Society<br />
meeting in 1947 (Koopmans, 1948).<br />
Dantzig (1949) stated the LP problem, not yet in standard format, but the<br />
solution technique was not discussed. The paper referred not only Leontief’s<br />
input-output model but also to John von Neumann’s work on economic equilibrium<br />
growth, i.e., to recent papers firmly within the realm of economics.<br />
Wood and Dantzig (1949) in the same issue stated an airlift problem.<br />
In 1949, Cowles Commission and RAND jointly arranged a conference<br />
on LP. The conference meant a breakthrough for the new optimization<br />
technique and had prominent participation. At the conference were<br />
economists, Tjalling Koopmans, Paul Samuelson, Kenneth Arrow, Leonid<br />
Hurwicz, Robert Dorfman, Abba Lerner and Nicholas Georgescu-Roegen,<br />
mathematicians, Al Tucker, Harold Kuhn and David Gale, and several military<br />
researchers including Dantzig.<br />
Dantzig had discovered the Simplex method but admitted many years<br />
later that he had not really realized how important this discovery was. Few<br />
people had a proper overview of linear models to place the new discovery in
8 Olav Bjerkholt<br />
context, but one of the few was John von Neumann, at the time an authority<br />
on a wide range of problems form nuclear physics to the development<br />
of computers. Dantzig decided to consult him about his work on solution<br />
techniques for the LP problem:<br />
“I decided to consult with the “great” Johnny von Neumann to see what<br />
he could suggest in the way of solution techniques. He was considered<br />
by many as the leading mathematician in the world. On October 3, 1947<br />
I visited him for the first time at the Institute for Advanced Study at<br />
Princeton. I remember trying to describe to von Neumann, as I would to<br />
an ordinary mortal, the Air Force problem. I began with the formulation<br />
of the linear programming model in terms of activities and items, etc. Von<br />
Neumann did something, which I believe was uncharacteristic of him.<br />
“Get to the point,” he said impatiently. Having at times a somewhat low<br />
kindling point, I said to myself “O.K., if he wants a quicky, then that’s what<br />
he’ll get.” In under one minute I slapped the geometric and the algebraic<br />
version of the problem on the blackboard. Von Neumann stood up and<br />
said “Oh that!” Then for the next hour and a half, he proceeded to give<br />
me a lecture on the mathematical theory of linear programs.” (Dantzig,<br />
1984).<br />
Von Neumann could immediately recognize the core issue as he saw the<br />
similarity with the game theory. The meeting became the first time Dantzig<br />
heard about duality and Farkas’ lemma.<br />
One could underline how LP once seemed firmly anchored in economics<br />
by pointing to the number of Nobel Laureates in economics who<br />
undertook studies involving LP. These comprise Samuelson and Solow, coauthors<br />
with Dorfman of Dorfman, Samuelson and Solow (1958), Leontief<br />
who applied LP in his input-output models, and Koopmans and Stigler,<br />
originators of the transportation problem and the diet problem, respectively.<br />
One also has to include Frisch, as we shall look at in more detail below,<br />
Arrow, co-author of Arrow et al. (1958), Kantorovich, who is known to<br />
have formulated the LP problem already in 1939, Modigliani, and Simon,<br />
who both were co-authors of Holt et al. (1956). Finally, to be included<br />
are also all the three Nobel Laureates in economics for 1990: Markowitz<br />
et al. (1957), Charnes et al. (1959) and Sharpe (1967). Koopmans and<br />
Kantorovich shared the Nobel Prize in economics for 1975 for their work<br />
in LP. John von Neumann died long before the Nobel Prize in economics<br />
was established, and George Dantzig never got the prize he ought to have<br />
shared with Koopmans and Kantorovich in 1975 for reasons that are not<br />
easy to understand.
Some Unresolved Problems 9<br />
Among all these Nobel Laureates, it seems that Ragnar Frisch had the<br />
deepest involvement with LP. But he published poorly and his achievements<br />
went largely unrecognized.<br />
3. Frisch and Programming, Pre-War Attempts<br />
Ragnar Frisch had a strong mathematical background and a passion for<br />
numerical analysis. He had been a co-founder of the Econometric Society<br />
in 1930 and to him econometrics meant both the formulation of economic<br />
theory in a precise way by means of mathematics and the development<br />
of methods for confronting theory with empirical data. He wrote both<br />
of these aims into the constitution of the Econometric Society as rendered<br />
for many years in every issue of Econometrica. He became editor of<br />
Econometrica from its first issue in 1933 and held this position for more than<br />
20 years.<br />
His impressive work in several fields of econometrics must be left<br />
uncommented here. Frisch pioneered the use of matrix algebra in econometrics<br />
in 1929 and had introduced many other innovations in the use of<br />
mathematics for economic analysis.<br />
Frisch’s most active and creative as a mathematical economist coincided<br />
with the Great Depression, which he observed at close range both<br />
in the United States and in Norway. In 1934, he published an article<br />
in Econometrica (Frisch, 1934a) where he discussed the organization of<br />
national exchange organization that could take over the need for trade<br />
when markets had collapsed due to the depression, see Bjerkholt and Knell<br />
(2006). In many ways, the article may seem naïve with regard to the subject<br />
matter, but it had a very sophisticated mathematical underpinning.<br />
The article was 93 pages long, the longest regular article ever published<br />
in Econometrica. Frisch had also published the article without letting it<br />
undergo proper referee process. Perhaps it was the urgency of getting it<br />
circulated that caused him to do this mistake for which he was rebuked<br />
by some of his fellow econometricians. The article was written as literally<br />
squeezed in between Frisch two most famous econometric contributions<br />
in the 1930s, his business cycle theory (Frisch, 1933) and his confluence<br />
theory (Frisch, 1934b).<br />
Frisch (1934a) developed a linear structure representing inputs for production<br />
in different industries, rather similar to the input-output model<br />
of Leontief, although the underlying idea and motivation was quite different.<br />
Frisch then addressed the question of how the input needs could
10 Olav Bjerkholt<br />
be modified in the least costly way in order to achieve a higher total<br />
production capacity of the entire economy. The variables in his problem<br />
(x i ) represented (relative) deviations from observed input coefficients.<br />
He formulated a cost function as a quadratic function of these deviations.<br />
His reasoning thus led to a quadratic target function to be minimized<br />
subject to a set of linear constraints. The problem was stated as<br />
follows:<br />
[ ]<br />
min = 1 x<br />
2<br />
1<br />
+ x2 2<br />
+···+ x2 n<br />
wrt. x k k = 1, 2,...,n<br />
2 ε 1 ε 2 ε n<br />
subject to: k f ik x k ≥ C ∗ − a i0<br />
i = 1,...,n<br />
P i<br />
where C ∗ is the set target of overall production and the constraint the condition<br />
that the deviation allowed this target value to be achieved.<br />
By solving for alternative values of C ∗ , the function = (C ∗ ),<br />
representing the cost of achieving the production level C ∗ could be mapped.<br />
Frisch naturally did not possess an algorithm for this problem. This did<br />
not prevent him from working out a numerical solution of an illuminating<br />
example and making various observations on the nature of such optimising<br />
problems. His solution of the problem for a given example with n = 3,<br />
determining as a function of C ∗ ran over 12 pages in Econometrica, see<br />
Bjerkholt (2006).<br />
Another problem Frisch worked on before the war with connections<br />
to LP was the diet problem. One of Frisch’s students worked on a dissertation<br />
on health and nutrition, drawing on various investigations of the<br />
nutritional value of alternative diets. Frisch supervised the dissertation in<br />
the years immediately before World War II when it led him to formulate<br />
the diet problem, years before Stigler (1945). Frisch’s work was published<br />
as the introduction to his student’s monograph (Frisch, 1941). In a footnote,<br />
he gave a somewhat rudimentary LP formulation of the diet problem,<br />
cf. Sandmo (1993).<br />
4. Frisch and Linear Programming<br />
Frisch was not present at the Econometric Society meetings in 1948 mentioned<br />
above; neither did he participate in the Cowles Commission-RAND<br />
conference in 1949. He could still survey and follow these developments<br />
rather closely. He was still the editor of Econometrica and thus had accepted<br />
and surely studied the two papers by Dantzig in 1949. He had contact with<br />
Cowles Commission in Chicago, its two successive research directors in
Some Unresolved Problems 11<br />
these years, Jacob Marschak and Tjalling Koopmans, and other leading<br />
members of the Econometric Society. Frisch visited in the early post-war<br />
years frequently in the United States, primarily due to his position as chairman<br />
of the UN Sub-Commission for Employment and <strong>Economic</strong> Stability.<br />
The new developments over the whole range of related areas of linear techniques<br />
had surely caught his interest.<br />
He first touched upon LP in lectures to his students at the University of<br />
Oslo in September 1950. Two students drafted lecture notes that Frisch corroborated<br />
and approved. In these early lectures Frisch used “input-output<br />
analysis” and “LP” as almost synonymous concepts. The content of these<br />
early lectures was input-output analysis with emphasis on optimisation by<br />
means of LP, e.g., optimisation under a given import constraint or given<br />
labor supply, they did not really address LP techniques as a separate issue.<br />
Still these lectures may well have been among the first applying LP to<br />
the macroeconomic problems of the day in Western Europe, years before<br />
Dorfman et al. (1958) popularised this tool for economists’ benefit. One<br />
major reason for Frisch pioneering efforts here was, of course, that the<br />
new ideas touched or even overlapped with his own pre-war attempts as<br />
discussed above. In the very first post-war years, he had indeed applied<br />
the ideas of Frisch (1934a) to the problem of achieving the largest possible<br />
multilateral balanced trade in the highly constrained world economy<br />
(Frisch, 1947; 1948).<br />
The first trace of LP as a topic in its own right on Frisch’s agenda<br />
was a single lecture he gave to his students on February 15th 1952. Leif<br />
Johansen, who was Frisch’s assistant at the time, took notes that were issued<br />
in the Memorandum series two days later (Johansen, 1952). The lecture<br />
gave a brief mathematical statement of the LP problem and then as an<br />
important example discussed the problem related to producing different<br />
qualities of airplane fuel from crude oil derivates, with reference to an<br />
article by Charnes et al. in Econometrica. Charnes et al. (1952) is, indeed,<br />
a famous contribution in the LP literature as the first published paper on<br />
an industrial application of LP. A special thing about the Frisch’s use of<br />
this example was, however, that his lecture did not refer the students to<br />
an article that had already appeared in Econometrica, but to one that was<br />
forthcoming! Frisch had apparently been so eager to present these ideas<br />
that he had simply used (or perhaps misused) his editorial prerogative in<br />
lecturing and issuing lecture notes on an article not yet published!<br />
After the singular lecture on LP in February 1952, Frisch did not lecture<br />
on this topic till the autumn term in 1953, as part of another lecture series<br />
on input-output analysis. In these lectures, he went by way of introduction
12 Olav Bjerkholt<br />
through a number of concrete examples of problem that were amenable to<br />
be solved by LP, most of which have already been mentioned above. When<br />
he came to the diet problem, it was not with reference to Stigler (1945), but<br />
to Frisch (1941). He exemplified, not least, with reconstruction in Norway,<br />
the building of dwellings under the tight post-war constraints and various<br />
macroeconomic problems.<br />
In his brief introduction to LP, Frisch in a splendid pedagogical way<br />
used examples, amply illustrated with illuminating figures, and appealed to<br />
geometric intuition. He started out with cost minimization of a potato and<br />
herring diet with minimum requirements of fat, proteins and vitamin B, and<br />
then moved to macroeconomic problems within an input-output framework,<br />
somewhat similar in tenor to Dorfman (1958), still in the offing.<br />
Frisch confined, however, his discussion to the formulation of the LP<br />
problem and the characteristics of the solution, he did not enter into solution<br />
techniques. But clearly he had already put considerable work into this.<br />
Because after mentioning Dantzig’s Simplex method, referring to Dantzig<br />
(1951), Dorfman (1951) and Charnes et al. (1953), he added that there<br />
were two other methods, the logarithm potential method and the basis<br />
variable method, both developed at the Institute of <strong>Economic</strong>s. He added<br />
the remark that when the number of constraints was moderate and thus the<br />
number of degrees of freedom large, the Simplex method might well be<br />
the most efficient one. But in other cases, the alternative methods would be<br />
more efficient. Needless to say, there were no other sources, which referred<br />
to these methods at this time. Frisch’s research work on LP methods and<br />
alternatives to the Simplex method, can be followed almost step by step<br />
thanks to his working style of incorporating new results in the Institute<br />
Memorandum series. Between the middle of October 1953 and May 1954,<br />
he issued 10 memoranda.They were all in Norwegian, but Frisch was clearly<br />
preparing for presenting his work. As were to be expected Frisch’s work on<br />
LP was sprinkled with new terms, some of them adapted from earlier works.<br />
Some of the memoranda comprised comprehensive numerical experiments.<br />
The memoranda were the following:<br />
18 October 1953 Logaritmepotensialmetoden til løsning av lineære<br />
programmeringsproblemer [The logarithm potential<br />
method for solving LP problems], 11 pages.<br />
7 November 1953 Litt analytisk geometri i flere dimensjoner [Some<br />
multi-dimensional analytic geometry], 11 pages.
Some Unresolved Problems 13<br />
(Continued )<br />
13 November 1953 Substitumalutforming av logaritmepotensialmetoden til<br />
løsning av lineære programmeringsproblemer<br />
[Substitumal formulation of the logarithmic potential<br />
method for solving LP problems], 3 pages.<br />
14 January 1954 Notater i forbindelse med logaritmepotensialmetoden til<br />
løsning av lineære programmeringsproblemer [Notes<br />
in connection with the logarithmic potential method<br />
for solving LP problems], 48 pages.<br />
7 March 1954 Finalhopp og substitumalfølging ved lineær<br />
programmering [Final jumps and moving along the<br />
substitumal in LP], 8 pages.<br />
29 March 1954 Trunkering som forberedelse til finalhopp ved lineær<br />
programmering [Truncation as preparation for a final<br />
jump in LP], 6 pages.<br />
27 April 1954 Et 22-variables eksempel på anvendelse av<br />
logaritmepotensialmetoden til løsning av lineære<br />
programmeringsproblemer [A 22 variable example in<br />
application of the logarithmic potential method for the<br />
solution of LP problems].<br />
1 May 1954 Basisvariabel-metoden til løsning av lineære<br />
programmeringsproblemer [The basis-variable<br />
method for solving LP problems], 21 pages.<br />
5 May 1954 Simplex-metoden til løsning av lineære<br />
programmeringsproblemer [The Simplex method for<br />
solving LP problems], 14 pages.<br />
7 May 1954 Generelle merknader om løsningsstrukturen ved det<br />
lineære programmeringsproblem [Generfal remarks<br />
on the solution structure of the LP problem],<br />
12 pages.<br />
At this stage, Frisch was ready to present his ideas internationally. This<br />
happened first at a conference in Varenna in July 1954, subsequently issued<br />
as a memorandum. During the summer, Frisch added another couple of<br />
memoranda. Then he went to India, invited by Pandit Nehru, and spent the<br />
entire academic year 1954–55 at the Indian Statistical Institute, directed<br />
by P. C. Mahalanobis, to take part in the preparation of the new five-year<br />
plan for India. While he was away, he sent home the MSS for three more
14 Olav Bjerkholt<br />
memoranda. These six papers were the following:<br />
June 21st 1954<br />
August 13th 1954<br />
August 23rd 1954<br />
October 18th 1954<br />
March 29th 1955<br />
April 15th 1955<br />
Methods of solving LP problems, Synopsis of lecture to be<br />
given at the International Seminar on Input-Output<br />
Analysis, Varenna (Lake Como), June–July 1954,<br />
91 pages.<br />
Nye notater i forbindelse med logaritmepotensialmetoden<br />
til løsning av lineære programmeringsproblemer [New<br />
notes in connection with the logarithmic potential<br />
method for solving LP problems], 74 pages.<br />
Merknad om formuleringen av det lineære<br />
programmeringsproblem [A remark on the formulation of<br />
the LP problem], 3 pages.<br />
Principles of LP. With particular reference to the double<br />
gradient form of the logarithmic potential method,<br />
219 pages.<br />
A labour saving method of performing freedom truncations<br />
in international trade. Part I, 21 pages.<br />
A labour saving method of performing freedom truncations<br />
in international trade. Part II, 8 pages.<br />
Before he left for India he had already accepted an invitation to present<br />
his ideas on programming in Paris. This he did on three different occasions<br />
in May–June 1955. He presented in French, but issued synopsis in English<br />
in the memorandum series. Thus, his Paris presentation were the following:<br />
7th May 1955 The logarithmic potential method for solving linear<br />
programming problems, 16 pp. Synopsis of an exposition to<br />
be made on 1 June 1955 in the seminar of Professor René<br />
Roy, Paris (published as Frisch, 1956b).<br />
13 May 1955 The logarithmic potential method of convex programming. With<br />
particular application to the dynamics of planning for<br />
national developments, 35 pp. Synopsis of a presentation to<br />
be made at the international colloquium in Paris, 23–28 May<br />
1955 (published as Frisch, 1956c).<br />
25 May 1955 Linear and convex programming problems studied by means of<br />
the double gradient form of the logarithmic potential method,<br />
16pp. Synopsis of a presentation to be given in the seminar of<br />
Professor Allais, Paris, 26 May 1955.
Some Unresolved Problems 15<br />
After this, Frisch published some additional memoranda on LP. He<br />
launched a new method, named the Multiplex method, towards the end of<br />
1955, later published in a long article in Sankhya (Frisch, 1955; 1957).<br />
Worth mentioning is also his contribution to the festschrift for Erik Lindahl<br />
in which he discussed the logarithmic potential method and linked it to the<br />
general problem of macroeconomic planning (Frisch, 1956a; 1956d).<br />
From this time, Frisch’s efforts subsided with regard to pure LP problems,<br />
as he concentrated on non-linear and more complex optimisation,<br />
drawing naturally on the methods he had developed for LP.<br />
As Frisch managed to publish his results only to a limited degree and<br />
not very prominently there are few traces of Frisch in the international<br />
literature. Dantzig (1963) has, however, references to Frisch’s work. Frisch<br />
also corresponded with Dantzig, Charnes, von Neumann and others about<br />
LP methods. Extracts from such correspondence are quoted in some of the<br />
memoranda.<br />
5. Frisch’s Radar: An Anticipation of Karmarkar’s Result<br />
In 1972, severe doubts were created about the ability of the Simplex method<br />
to solve even larger problems. Klee and Minty (1972), whom we must<br />
assume knew very well how the Simplex algorithm worked, constructed<br />
a LP problem, which when attacked by the Simplex method turned out<br />
to need a number of elementary operations which grew exponentially in<br />
the number of variables in the problem, i.e., the algorithm had exponential<br />
complexity. Exponential complexity in the algorithm is for obvious reasons<br />
a rather ruining property for attacking large problems.<br />
In practice, however, the Simplex method did not seem to confront such<br />
problems, but the demonstration of Klee and Minty (1972) showed a worst<br />
case complexity, implying that it could never be proved what had been<br />
largely assumed, namely that the Simplex method only had polynomial<br />
complexity.<br />
A response to Klee and Minty’s result came in 1978 by the Russian<br />
mathematician Khachiyan 1978 (published in English in 1979 and 1980).<br />
He used a so-called “ellipsoidal” method for solving the LP problem, a<br />
method developed in the 1970s for solving convex optimisation problems by<br />
the Russian mathematicians Shor and Yudin and Nemirovskii. Khachiyan<br />
managed to prove that this method could indeed solve LP problems and<br />
had polynomial complexity as the time needed to reach a solution increased<br />
with the forth power of the size of the problem. But the method itself was
16 Olav Bjerkholt<br />
hopelessly ineffective compared to the Simplex method and was in practice<br />
never used for LP problems.<br />
A few years later, Karmarkar presented his algorithm (Karmarkar,<br />
1984a; b). It has polynomial complexity, not only of a lower degree than<br />
Khachiyan’s method, but is asserted to be more effective than the Simplex<br />
method. The news about Karmarkar’s path-breaking result became frontpage<br />
news in the New York Times. For a while, there was some confusion as<br />
to whether Karmarkar’s method really was more effective than the Simplex<br />
method, partly because Karmarkar had used a somewhat special formulation<br />
of the LP problem. But it was soon confirmed that for sufficently large<br />
problems, the Simplex method was less effcient than Karmarkar’s result.<br />
Karmarkar’s contribution initiated a hectic new development towards even<br />
better methods.<br />
Karmarkar’s approach may be said to be to consider the LP just as<br />
another convex optimalization problem rather that exploiting the fact that<br />
the solution must be found in a corner by searching only corner points. One<br />
of those who has improved Karmarkar’s method, Clovis C. Gonzaga, and<br />
for a while was in the lead with regard to possessing the nest method wrt.<br />
polynomial complexity, said the following about Karmarkar’s approach:<br />
“Karmarkar’s algorithm performs well by avoiding the boundary of the<br />
feasible set. And it does this with the help of a classical resource, first<br />
used in optimization by Frisch in 1955: the logarithmic barrier function:<br />
x ∈ R n , x > 0 → p(x) =− ∑ log x i .<br />
This function grows indefinitely near the boundary of the feasible set S,<br />
and can be used as a penalty attached to these points. Combining p(·)<br />
with the objective makes points near the boundary expensive, and forces<br />
any minimization algorithm to avoid them.” (Gonzaga, 1992)<br />
Gonzaga’s reference to Frisch was surprising; it was to the not very<br />
accessible memorandum version in English of one of the three Paris papers. 1<br />
Frisch had indeed introduced the logarithmic barrier function in his logarithmic<br />
potential method. Frisch’s approach was summarized by Koenker<br />
as follows:<br />
“The basic idea of Frisch (1956b) was to replace the linear inequality<br />
constraints of the LP, by what he called a log barrier, or potential function.<br />
Thus, in place of the canonical linear program<br />
1 In Gonzaga’s reference, the author’s name was given as K.R. Frisch, which can also be<br />
found in other rerences to Frisch in recent literature, suggesting that these refrences had a<br />
common source.
(1) min{c ′ x|Ax = b, x ≥ 0}<br />
we may associate the logarithmic barrier reformulation<br />
(2) min{B(x, µ)|Ax = b}<br />
where<br />
(3) B(x, µ) = c ′ x − µ ∑ log x k<br />
Some Unresolved Problems 17<br />
In effect, (2) replaces the inequality constraints in (1) by the penalty term<br />
of the log barrier. Solving (2) with a sequence of parameters µ such that<br />
µ → 0 we obtain in the limit a solution to the original problem (1).”<br />
(Koenker, 2000, p. 20)<br />
Frisch had not provided exact proofs and he had above all not published<br />
properly. But he had the exact same idea as Karmarkar came up with almost<br />
30 years later. Why did he not make his results better known? In fact there<br />
are other, even more important cases, in Frisch’s work of not publishing. The<br />
reasons for this might be that his investigations had not yet come to an end.<br />
In the programming work Frisch pushed on to more complex programming<br />
problems, spurred by the possibility of using first- and second-generation<br />
computers. They might have seemed powerful to him at that time, but they<br />
hardly had the capacity to match Frisch’s ambitions. Another reason for<br />
his results remaining largely unknown, was perhaps that he was too much<br />
ahead. The Simplex method had not really been challenged yet, by large<br />
enough problems to necessitate better method. The problem may have seen,<br />
not as much as a question of efficient algorithms as that of powerful enough<br />
computers.<br />
We finish off with Frisch’s nice illustrative description of his method in<br />
the only publication he got properly published (but unfortunately not very<br />
well distributed) on the logarithmic potential method:<br />
“Ma méthode d’approche est d’une espèce toute différente. Dans cette<br />
méthode nous travaillons systématiquement à l’intérieur de la région<br />
admissible et utilisons un potentiel logarithmique comme un guide–<br />
une sorte du radar–pour nous éviter de traverser la limite.” (Frisch,<br />
1956b, p. 13)<br />
The corresponding quote in the memorandum synoptic note is:<br />
“My method of approach is of an entirely different sort [than the Simplex<br />
method]. In this method we work systematically in the interior of the<br />
admissible region and use a logarithmic potential as a guiding device — a<br />
sort of radar — to prevent us from hitting the boundary.” (Memorandum<br />
of 7 May 1955, p. 8)
18 Olav Bjerkholt<br />
References<br />
Arrow, KJ, L Hurwicz and H Uzawa (1958). Studies in Linear and Non-Linear Programming.<br />
Stanford, CA: Stanford University Press.<br />
Bjerkholt, O and M Knell (2006). Ragnar Frisch and input-output analysis. <strong>Economic</strong><br />
Systems Research 18, (forthcoming).<br />
Charnes, A, WW Cooper and A Henderson (1953). An Introduction to Linear Programming.<br />
New York and London: John Wiley & Sons.<br />
Charnes, A, WW Cooper and B Mellon (1952). Bledning aviation gasolines — a study in<br />
programming interdependent activities in an integrated oil company, Econometrica 20,<br />
135–159.<br />
Charnes, A, WW Cooper and MW Miller (1959). An application of linear programming to<br />
financial budgeting and the costing of funds. The Journal of Business 20, 20–46.<br />
Dantzig, GB (1949). Programming of interdependent activities: II mathematical model.<br />
Econometrica 17, 200–211.<br />
Dantzig, GB (1951). Maximization of a linear function of variables subject to linear inequalities.<br />
In Activity Analysis of Production and Allocation, T Koopmans (ed.), New York<br />
and London: John Wiley & Sons, pp. 339–347.<br />
Dantzig, GB (1963). Linear Programming and Extensions. Princeton, NJ: Princeton<br />
University Press.<br />
Dantzig, GB (1984). Reminiscences about the origin of linear programming. In<br />
Mathematical Programming, RW Cottle, ML Kelmanson and B Korte (eds.),<br />
Amsterdam: Elsevier (North-Holland), pp. 217–226.<br />
Dorfman, R (1951). An Application of Linear Programming to the Theory of the Firm.<br />
Berkeley: University of California Press.<br />
Dorfman, R, PA Samuelson and RM Solow, (1958). Linear Programming and <strong>Economic</strong><br />
Analysis. New York: McGraw-Hill.<br />
Frisch, R (1933). Propagation problems and impulse problems in economics. In<br />
<strong>Economic</strong> Essays in Honor of Gustav Cassel, R Frisch (ed.), London: Allen & Unwin,<br />
pp. 171–205.<br />
Frisch, R (1934a). Circulation planning: proposal for a national organization of a commodity<br />
and service exchange. Econometrica 2, 258–336 and 422–435.<br />
Frisch, R (1934b). Statistical Confluence Analysis by Means of Complete Regression<br />
Systems. Publikasjon nr 5, Oslo: Institute of <strong>Economic</strong>s.<br />
Frisch, R (1941). Innledning. In Kosthold og levestandard. En økonomisk undersøkelse,<br />
K Getz Wold (ed.), Oslo: Fabritius og Sønners Forlag, pp. 1–23.<br />
Frisch, R (1947). On the need for forecasting a multilateral balance of payments. The American<br />
<strong>Economic</strong> Review 37(1), 535–551.<br />
Frisch, R (1948). The problem of multicompensatory trade. Outline of a system of multicompensatory<br />
trade. The Review of <strong>Economic</strong>s and Statistics 30, 265–271.<br />
Frisch, R (1955). The multiplex method for linear programming. Memorandum from Institute<br />
of <strong>Economic</strong>s, Oslo: University of Oslo (17 October 1955).<br />
Frisch, R (1956a). Macroeconomics and linear programming. Memorandum from Institute<br />
of <strong>Economic</strong>s, Oslo: University of Oslo, (10 January 1956). (Also published shortened<br />
and simplified as Frisch (1956d)).<br />
Frisch, R (1956b). La résolution des problèmes de programme linéaire par la méthode du<br />
potential logarithmique. In Cahiers du Séminaire d’Econometrie: No4—Programme
Some Unresolved Problems 19<br />
Linéaire — Agrégation et Nombre Indices, pp. 7–20. (The publication also gives an<br />
extract of the oral discussion in René Roy’s seminar, pp. 20–23.)<br />
Frisch, R (1956c). Programme convexe et planification pour le développement national.<br />
Colloques Internationaux du Centre National de la Recherche Scientifique LXII. Les<br />
modèles dynamiques. Conférence à Paris, mai 1955, 47–80.<br />
Frisch, R (1956d). Macroeconomics and linear programming. In 25 <strong>Economic</strong> Essays. In<br />
honour of Erik Lindahl 21 November 1956, R Frisch (ed.), pp. 38–67. Stockholm:<br />
Ekonomisk Tidskrift, (Slightly simplified version of Frisch (1956a)).<br />
Frisch, R (1957). The multiplex method for linear programming. Sankhyá: The Indian<br />
Journal of Statistics 18, 329–362 [Practically identical to Frisch (1955)]<br />
Gonzaga, CC (1992). Path-following methods for linear programming. SIAM Review 34,<br />
167–224.<br />
Holt, CC, F Modigliani, JF Muth and HA Simon (1956). Derivation of a linear decision rule<br />
for production and employment. Management Science 2, 159–177.<br />
Hurwicz, L (1949). Linear programming and general theory of optimal behaviour (abstract).<br />
Econometrica 17, 161–162.<br />
Johansen, L (1952). Lineær programming. (Notes from Professor Ragnar Frisch’s lecture<br />
15 February 1952), pp. 15. Memorandum from Institute of <strong>Economic</strong>s, University of<br />
Oslo, 17 February 1952, 15pp.<br />
Karmarkar, NK (1984). A new polynomial-time algorithm for linear programming. Combinatorica,<br />
4, 373–395.<br />
Klee, V and G Minty, (1972). How good is the simplex algorithm. In Inequalities III, O<br />
Sisha (ed.), pp. 159–175. New York, NY: Academic Press.<br />
Koenker, R (2000). Galton, Edgeworth, Frisch, and prospects for quantile regression in<br />
econometrics. Journal of Econometrics, 95(2), 347–374.<br />
Kohli, MC (2001). Leontief and the bureau of labor statistics, 1941–54: developing a<br />
framework for measurement. In The Age of <strong>Economic</strong> Measurement, JL Klein and<br />
MS Morgan (eds.), pp. 190–212. Durham and London: Duke University Press.<br />
Koopmans, T (1948). Optimum utilization of the transportation system (abstract).<br />
Econometrica 16, 66.<br />
Koopmans, TC (ed.) (1951). Activity Analysis of Production and Allocation. Cowles<br />
Commission Monographs No 13. New York: John Wiley & Sons.<br />
Markowitz, H (1957). The elimination form of the inverse and its application in linear<br />
programming. Management Science 3, 255–269.<br />
Orden, A (1993). LP from the ’40s to the ’90s. Interfaces 23, 2–12.<br />
Sandmo, A (1993). Ragnar Frisch on the optimal diet. History of Political Economy 25,<br />
313–327.<br />
Sharpe, W (1967). A linear programming algorithm for mutual fund portfolio selection.<br />
Management Science 13, 499–511.<br />
Stigler, G (1945). The cost of subsistence. Journal of Farm <strong>Economic</strong>s 27, 303–314.<br />
Wood, MK and GB Dantzig (1949). Programming of interdependent activities: I general<br />
discussion. Econometrica 17, 193–199.
This page intentionally left blank
Chapter 2<br />
Methods of Modeling: Econometrics<br />
and Adaptive Control System
This page intentionally left blank
Topic 1<br />
A Novel Method of Estimation<br />
Under Co-Integration<br />
Alexis Lazaridis<br />
Aristotle University of Thessaloniki, Greece<br />
1. Introduction<br />
We start with the unrestricted VAR(p):<br />
p∑<br />
x i = δ + µt i + A j x i−j + zd i + w i (1)<br />
j=1<br />
where x ∈ E n (here, E denotes the Euclidean space), t i and d i are the<br />
trend and dummy, respectively and w is the noise vector. The matrices of<br />
coefficients A j , are defined on E n × E n . After estimation, we may compute<br />
matrix from:<br />
⎛ ⎞<br />
p∑<br />
= ⎝ A j<br />
⎠ − I<br />
(1a)<br />
j=1<br />
or, directly, from the corresponding Error CorrectionVAR (ECVAR), which<br />
is presented later.Also, an estimate of matrix can be obtained by applying<br />
OLS, according to the procedure suggested by Kleibergen and Van Dijk<br />
(1994). It is re-called that this type of VAR models like the one presented<br />
above, have been recommended by Sims (1980), as a way to estimate the<br />
dynamic relationships among jointly endogenous variables.<br />
It is known that matrix can be decomposed such that = AC,<br />
where the elements of A are the coefficients of adjustment and the rows of<br />
the matrix C are the (possible) co-integrating vectors. It should be clarified<br />
at the very outset, that in many books etc., C is denoted by β ′ (prime shows<br />
transposition) and A by α, although capital letters are used for denoting<br />
matrices. Besides, in the general linear model, y = Xβ + u,β is the vector<br />
of coefficients. To avoid any confusion, we adopted the notation used here.<br />
23
24 Alexis Lazaridis<br />
A straightforward approach to obtain matrices A and C, provided that<br />
some rank conditions of are satisfied, is to solve the eigenproblem (Johnston<br />
and Di Nardo, 1997, pp. 287–296).<br />
= VV −1 (2)<br />
where is the diagonal matrix of eigenvalues and V is the matrix of the<br />
corresponding eigenvectors. Note that since is not symmetric, V ̸= V −1 .<br />
Eq. (2) holds if is (n × n) and all its eigenvalues are real, which implies<br />
that the eigenvectors are real too. In cases that is defined on E n × E m ,<br />
with m>n, Eq. (2) is not applicable. If these necessary conditions hold,<br />
then from Eq. (2) follows that matrix A = V and C = V −1 . Further, the<br />
co-integrating vectors are normalized so that — usually — the first element<br />
to be equal to 1. In other words, if c i 1 (̸=0), denotes the first element1 of<br />
the ith row of matrix C, i.e., 2 c ′ i. and a j the jth column of matrix A, then<br />
the normalization process can be summarized as c ′ i. /ci 1 and a i × c i 1 , for<br />
i = 1,...,n.<br />
It is known that according to the maximum likelihood (ML) method,<br />
matrix C can be obtained by solving the constrained problem:<br />
{min det(I − C ˆ k0 ˆ −1<br />
00 ˆ 0k C ′ ) | C ˆ kk C ′ = I} (3)<br />
where matrices ˆ 00 , ˆ 0k , ˆ kk and ˆ k0 are residual co-variance matrices of<br />
specific regressions (Harris, 1995, p. 78; Johansen Juselius, 1990). These<br />
matrices appear in the last term of the concentrated likelihood of 3 Eq. (5)<br />
together with C, where is replaced by AC.<br />
Note that the minimization of Eq. (3) is with respect to the elements<br />
of matrix C. It is re-called that Eq. (3) is minimized by solving a general<br />
eigenvalue problem, i.e., finding the eigenvalues from:<br />
|λ ˆ kk − ˆ k0 ˆ −1<br />
00 ˆ 0k |=0. (4)<br />
Applying Cholesky’s factorization on the positive definite symmetric<br />
matrix ˆ −1<br />
kk , such that ˆ −1<br />
kk = LL′ ⇒ ˆ kk = (L ′ ) −1 L −1 , where L is lower<br />
1 In some computer programs, ci i is taken to be the ith element of the ith row.<br />
2 Note that dot is necessary to distinguish the jth row of C, i.e., c i. ′ , from the transposed of<br />
the jth column of this matrix (i.e., c i ′).<br />
3 Equation (5) is presented below.
A Novel Method of Estimation Under Co-Integration 25<br />
triangular, Eq. (4) takes the conventional form 4 :<br />
|λI − L ′ ˆ k0 ˆ −1<br />
00 ˆ 0k L|=0. (4a)<br />
After obtaining matrix V = V ′ of the eigenvectors, we can easily<br />
compute the (unormalized) C from C = V ′ L ′ . Note also that C ˆ kk C ′ =<br />
V ′ L ′ (L ′ ) −1 L −1 LV = I. Finally, matrix A is obtained from A = ˆ 0k C ′ .<br />
Hence, the initial problem reduces to finding one of the eigen values and<br />
eigenvectors of a symmetric matrix, which are real. And this is the main<br />
advantage of the ML method.According to this procedure, a constant and/or<br />
trend can be included in the co-integrating vectors, but it is not clear of how<br />
to accommodate additional deterministic factors, such as one or more dummies.<br />
Besides, nothing is said about the variance of the (disequilibrium)<br />
errors, which correspond to the finally adopted co-integration vector.<br />
We propose in the first part of this chapter, a comparatively simple<br />
method, to directly obtain co-integrating vectors, which may include, apart<br />
from the standard deterministic components, any number of dummies, in<br />
cases that this is necessary. Additionally, as we will see in what follows, the<br />
errors corresponding to the proper co-integration vector have the property<br />
of minimum variance.<br />
2. Some Estimation Remarks<br />
It was mentioned that after estimation, matrix can be computed from<br />
Eq. (1a). It is obvious that each equation of the VAR can be estimated by<br />
OLS. At this stage, one should be aware to comply with the basic assumptions<br />
regarding the error terms, in the sense that proper tests must ensure<br />
that the noises are white. This way, the order p of the model is defined<br />
imposing at the same time restrictions — if necessary — on some elements<br />
of A’s and vectors µ and z to be zero.<br />
It can be easily shown that Eq. (1) may be expressed in the form of an<br />
equivalent ECVAR, i.e.,<br />
p−1<br />
∑<br />
x i = δ + µt i + Q j x i−j + x i−p + zd i + w i (5)<br />
j=1<br />
4 It should be noted that it is also necessary to pre-multiply the elements of Eq. (4) by L ′<br />
and post-multiply them by L.
26 Alexis Lazaridis<br />
where<br />
⎛ ⎞<br />
j∑<br />
Q j = ⎝ A k − I⎠ .<br />
k=1<br />
3. The Singular Value Decomposition<br />
Matrix in Eq. (5) may be augmented to accommodate, as an additional<br />
column, the vector δ, µ or z. It is also possible to accommodate all these<br />
vectors in the following way:<br />
˜ = [ .δµz ] (6)<br />
In this case, ˜ is defined on E n × E m , with m>n.<br />
For conformability, the vector x i−p in Eq. (5) should be augmented<br />
accordingly by using the linear advance operator L (such that L k y i = y i+k ),<br />
i.e.,<br />
⎡ ⎤<br />
x i−p<br />
1<br />
˜x i−p = ⎢<br />
⎣ L p ⎥<br />
t i−p ⎦ .<br />
L p d i−p<br />
Hence, Eq. (5) takes the form:<br />
p−1<br />
∑<br />
x i = Q j x i−j + ˜ ˜X i−p + w i . (7)<br />
j=1<br />
We can compute the unique generalized inverse (or pseudo-inverse) of ˜,<br />
denoted by ˜ + (Lazaridis and Basu, 1981), from:<br />
˜ + = VF ∗ U ′ (8)<br />
where<br />
V is of dimension (m×m), and its columns are the orthonormal eigenvectors<br />
of ˜ ′ ˜.<br />
U is (n × m), and its columns are the orthonormal eigenvectors of ˜ ˜ ′ ,<br />
corresponding to the largest m eigenvalues of this matrix so that:<br />
U ′ U = V ′ V = VV ′ = I m (9)<br />
F is diagonal (m × m), with elements f ii f i , called the singular values<br />
of ˜, which are the non-negative square roots of the eigenvalues of ˜ ′ ˜.
If the rank of ˜ is k, i.e.,<br />
A Novel Method of Estimation Under Co-Integration 27<br />
r( ˜) = k ≤ n, then f 1 ≥ f 2 ≥···≥f k > 0<br />
and f k+1 = f k+2 =···=f m = 0<br />
F ∗ is diagonal (m × m), and fii ∗ f i ∗ = 1/f i .<br />
It should be noted that all the above matrices are real. Also note that if<br />
˜ has full row rank, then ˜ + is the right inverse of ˜, i.e., ˜ ˜ + = I n<br />
Hence, the singular value decomposition (SVD) of ˜ is:<br />
˜ = UFV ′ . (10)<br />
4. Computing the Co-Integrating Vectors<br />
We proceed to form a matrix F 1 of dimension (m × n) such that:<br />
f 1(i,j) =<br />
{ 0, if i ̸= j<br />
√<br />
fi , if i = j<br />
(11)<br />
In accordance to Eq. (11), we next form a matrix F 2 of dimension (n × m).<br />
It is verified that F 1 F 2 = F. Hence, Eq. (10) can be written as:<br />
˜ = AC<br />
where A = UF 1 of dimension (n×n) and C = F 2 V of dimension (n×m).<br />
After normalization, we get the possible co-integrating vectors as the<br />
rows of matrix C, whereas the elements of A can be viewed as approximations<br />
to the corresponding coefficients of adjustment.<br />
It is apparent that the co-integrating vectors can always be computed,<br />
given that r( ˜) > 0. Needless to say that the same steps are followed if ,<br />
instead of ˜, is considered.<br />
5. Case Study 1: Comparing Co-Integrating Vectors<br />
We first consider the VAR(2) model, presented by Holden and Peaman<br />
(1994). The x ∼I(1) vector consists of the variables C (consumption), I<br />
(income) and W (wealth), so that x ∈ E 3 . We used the same set of data<br />
presented at the end of this book (Table D3, pp. 196–198). To simplify<br />
the presentation, we did not consider the three dummies DD682, DD792,<br />
and DD883. Hence, the VAR in this case has a form analogous to Eq. (1),<br />
without trend and dummy components.
28 Alexis Lazaridis<br />
Estimating the equations of this VAR by OLS, we get the following<br />
results.<br />
⎡<br />
⎤ ⎡<br />
⎤<br />
0.059472 −0.073043 0.0072427<br />
0.064873<br />
⎢<br />
⎥ ⎢<br />
⎥<br />
= ⎣0.5486237 −0.5244858 −0.028041 ⎦ , δ = ⎣ 0.16781451⎦ .<br />
0.3595042 −0.2938160 −0.039400 −0.1554421<br />
(12)<br />
Hat (ˆ) is omitted for simplicity. One possible co-integration vector with<br />
intercept, obtained by applying the ML method is:<br />
1 − 0.9574698 − 0.048531 + 0.2912925 (13)<br />
and corresponds to the long-run relationship:<br />
C i = 0.9574698I i + 0.048531W i − 0.2912925.<br />
For the vector in Eq. (13) to be a co-integration vector, the necessary and<br />
sufficient condition is that the (disequilibrium) errors û i computed from:<br />
û i = C i − 0.9574698I i − 0.048531W i + 0.2912925 (14)<br />
to be stationary, i.e., {û i }∼I(0), since x ∼I(1).<br />
For distinct purposes, these ML errors obtained from Eq. (14) will be<br />
denoted by uml i .<br />
Applying SVD on ˜ = [.δ] from Eq. (12), we get matrix C which is:<br />
⎡<br />
⎤<br />
1 −0.9206225 −0.06545528 0.1110597<br />
⎢<br />
⎥<br />
C = ⎣ 1 0.3586208 −0.6305243 −6.403011 ⎦<br />
1 1.283901 −2.050713 0.4300255<br />
⎛ ⎞<br />
1.365344<br />
⎜ ⎟<br />
Euclidean norm = ⎝ 6.521098 ⎠ .<br />
2.653064<br />
In the column at the right-hand side, the (Euclidean) norm of each row<br />
of C is presented. It is noted also that the singular values of are:<br />
f 1 = 0.8979247, f 2 = 0.2304381 and f 3 = 0.0083471695.<br />
All f i are less than 1, indicating thus that at least one row of C can be<br />
regarded as a possible co-integration vector, in the sense that the resulting<br />
errors are likely to be stationary. This row usually has the smallest norm,
A Novel Method of Estimation Under Co-Integration 29<br />
which in this case, is the first one. Hence, the errors corresponding to this<br />
co-integration vector are obtained from:<br />
û i = C i − 0.9206225I i − 0.06545528W i + 0.1110597. (15)<br />
These errors will be denoted by usvd i . It should be pointed out here<br />
that the smallest norm condition is just an indication as to which row of<br />
matrix C to be considered first.<br />
It may be useful to re-call at this point that the co-integration vector<br />
reported by the authors (p. 111), which is obtained by applying OLS after<br />
omitting the first two observations from the initial sample is:<br />
1 − 0.9109 − 0.0790 + 0.1778.<br />
It should be clarified that this vector has been obtained by applying OLS<br />
to estimate directly the long-run relationship. Repeating this procedure,<br />
we get:<br />
1 − 0.910971 − 0.079761 + 0.178455.<br />
Hence, the errors are computed from:<br />
û i = C i − 0.910971I i − 0.079761W i + 0.178455. (16)<br />
These errors will be denoted by ubook i .<br />
The next task is to compare the above three series of errors, having a<br />
first look at the corresponding graphs, which are shown in Fig. 1.<br />
Although by a first glance the three series seem to be stationary, we shall<br />
go a bit further to see their properties in detail. Before going to the next<br />
section, it is worth to mention here that for the three series, the normality<br />
test using the Jarque-Bera criterion favors the null.<br />
Figure 1. The series {uml i }, {usvd i }, and {ubook i }.
30 Alexis Lazaridis<br />
5.1. Testing for Co-integration<br />
Since for the first two cases, ∑ û i ̸= 0, we estimated the following regression:<br />
q∑<br />
û i = b 1 + b 2 û i−1 + b j+2 û i−j + ε i . (17)<br />
The value of q was set such that the noises ε i to be white, as it will be<br />
explained in detail later on. Regarding the last case, the intercept is omitted<br />
from Eq. (17), since û i ’s are OLS residuals, so that ∑ û i = 0.<br />
The estimation results are as follows:<br />
u ˆml i =−0.00612 − 0.381773 uml i−1 − 0.281651uml i−1<br />
(0.193688)<br />
j=1<br />
t =−3.682<br />
uŝvd i = 0.006525 − 0.428278 usvd i−1 − 0.259066usvd i−1<br />
(0.108269)<br />
t =−3.956<br />
ubôok i =−0.606759 ubook i−1<br />
(0.09522)<br />
t =−6.37.<br />
Since these error series are the results of specific calculations and in<br />
the simplest case are the OLS residuals, it is not advisable (Harris, 1995,<br />
pp. 54–55). to apply the Dickey-Fuller (DF/ADF) test, as we do with any<br />
variable in the initial data set. We have to compute the t u statistic from McKinnon<br />
(1991, p. 267–276) critical values (see also Harris, 1995, Table A6,<br />
p. 158). This statistic (t u = ∞ + 1 /T + 2 /T 2 ) for three (independent)<br />
variables in the long-run relationship with intercept is:<br />
α = 0.01 α = 0.05 α = 0.10<br />
t u =−4.45 t u =−3.83 t u =−3.52.<br />
Hence, {book i } is stationary for α = (0.01, 0.05, 0.10), {usvd i } is stationary<br />
for α = (0.05, 0.10) and {uml i } is stationary only for α = 0.10.<br />
Recall that the null [û i ∼ I(1)] is rejected in favor of H 1 [û i ∼ I(0)], if<br />
t
A Novel Method of Estimation Under Co-Integration 31<br />
It is obvious from these findings, the super consistency of the OLS.<br />
However, the method we introduce here, produces almost minimum variance<br />
errors, which is a quite desirable property, as suggested by Engle and<br />
Granger (Dickey et al., 1994, p. 27). Hence, we will mainly focus on this<br />
property in our next section.<br />
6. Case Study 2: Further Evidence Regarding the Minimum<br />
Variance Property<br />
With the same set of data, the co-integration vector without an intercept<br />
obtained by applying the ML method is:<br />
1 − 0.9422994 − 0.058564<br />
The errors corresponding to this vector are obtained from:<br />
û i = C i − 0.9422994I i − 0.058564W i (18)<br />
Again these errors will be denoted by uml i .<br />
Next, we apply SVD to in Eq. (12) to obtain matrix C, which is:<br />
⎡<br />
⎤<br />
1 −0.9192042 −0.066081211<br />
⎢<br />
⎥<br />
C = ⎣ 1 1.160720 −1.012970 ⎦<br />
1 0.9395341 2.063768<br />
and<br />
⎛ ⎞<br />
1.359891<br />
⎜ ⎟<br />
Euclidean norm = ⎝ 1.836676 ⎠<br />
2.47828<br />
The singular values of are:<br />
f 1 = 0.89514, f 2 = 0.0403121 and f 3 = 0.0026218249<br />
Again, all f i ’s are less than 1. We consider the first row of C, so that<br />
the errors corresponding to this vector are obtained from:<br />
û i = C i − 0.9192042I i − 0.066081211W i (19)<br />
These errors will be also denoted by usvd i .<br />
The co-integrating vector reported by the authors (p. 109), which is<br />
obtained from the VAR(2), including the dummies mentioned previously,
32 Alexis Lazaridis<br />
by applying the ML procedure is given by:<br />
1 − 0.93638 − 0.03804<br />
and the errors obtained from this vector are computed from:<br />
û i = C i − 0.93638I i − 0.03804W i (20)<br />
As in the previous case, these errors will be denoted by ubook i .<br />
The minimum variance property can be seen from the following:<br />
Standard deviation of Standard deviation of Standard deviation of<br />
{uml i } {usvd i } {book i }<br />
0.0169 0.0166 0.0193<br />
Next, we consider the model presented by Banerjee et al. (1993,<br />
pp. 292–293) that refers to VAR(2) with constant dummy and trend. The<br />
state vector x ∈ E 4 consists of the variables (m − p), p, y, and R. The<br />
data (in logs) are presented in Harris (1995, statistical appendix, pp. 153–<br />
155. Note that the entries for 1967:1 and 1973:2 have to be corrected from<br />
0.076195 and 0.56569498 to 9.076195 and 9.56569498, respectively). Considering<br />
the co-integration vector reported by the authors, which has been<br />
obtained by applying the ML method, we compute the errors from:<br />
û i = (m − p) i + 6.3966p i − 0.8938y i + 7.6838R i . (21)<br />
These errors will be denoted by uml i .<br />
Estimating the same VAR by OLS, we get,<br />
⎡<br />
⎤<br />
−0.089584 0.16375 −0.184282 −0.933878<br />
−0.012116 −0.9605 0.229635 0.206963<br />
= ⎢<br />
⎥<br />
⎣ 0.021703 0.676612 −0.476152 0.447157 ⎦ ,<br />
−0.006877 0.118891 0.048932 −0.076414<br />
⎡ ⎤<br />
⎡ ⎤<br />
3.129631<br />
0.001867<br />
−2.46328<br />
δ = ⎢ ⎥<br />
⎣ 5.181506 ⎦ and µ =<br />
−0.001442<br />
⎢ ⎥<br />
⎣ 0.003078 ⎦<br />
−0.470803<br />
−0.000366
A Novel Method of Estimation Under Co-Integration 33<br />
Following the procedure described above, we obtain matrix C which is:<br />
⎡<br />
⎤<br />
1 −48.30517 24.60144 37.75685<br />
1 5.491415 −0.6510755 7.423318<br />
C = ⎢<br />
⎥<br />
⎣ 1 −2.055229 −5.466846 0.9061699 ⎦<br />
1 0.0051181894 0.1603713 −0.1244312<br />
⎛ ⎞<br />
66.06966<br />
9.310488<br />
Euclidean norm = ⎜ ⎟<br />
⎝ 5.99429 ⎠<br />
1.020406<br />
The singular values of are:<br />
f 1 = 1.503069, f 2 = 0.7752298, f 3 = 0.1977592<br />
and f 4 = 0.012560162<br />
The fact that one singular value is greater than one, is an indication that<br />
unless a dummy is present — we hardly can trace one row of matrix C,<br />
which may be related to stationary errors û i . As shown in Fig. 1, from the<br />
graphs, from which refer to the errors corresponding to the rows of C, we<br />
see that only the second row produces reasonably results. In this case, the<br />
errors are computed from:<br />
û i = (m − p) i + 5.491415p i − 0.6510755y i + 7.42318R i . (22)<br />
As usual, these errors will be denoted by usvd i .<br />
The minimum variance property can be seen from:<br />
Standard deviation of {uml i } Standard deviation of {usvd i }<br />
0.2671 0.2375<br />
To see that none of the two error series is stationary, as it was indicated<br />
by the first singular value, we present the following results.<br />
u ˆml i = 0.268599 − 0.178771 uml i−1 − 0.214098uml i−1<br />
(0.065634)<br />
t =−2.724<br />
uŝvd i = 0.832329 − 0.193432 usvd i−1 − 0.160135usvd i−1<br />
(0.066421)<br />
t =−2.912<br />
For a long-run relationship with three (independent) variables without<br />
trend and intercept, the critical value t u for α = 0.10 is less than −3.
34 Alexis Lazaridis<br />
Nevertheless, the method proposed, produces better results than the ML<br />
approach does.<br />
Finally, we will consider the first of the two co-integration vectors<br />
(since, it produces slightly better results), reported by Harris (1995, p. 102).<br />
From this vector, which was obtained by the ML method, one gets the<br />
following errors.<br />
û i = (m − p) i + 7.325p i − 1.073y i + 6.808R i (23)<br />
Also we obtain<br />
standard deviation of û i ’s = 0.268 and<br />
û i =−0.117378 − 0.178896<br />
(0.068296) ûi−1 − 0.30095417û i−1<br />
t =−2.619.<br />
It is obvious that these additional results further support the previous<br />
findings, which are due to some properties of SVD (Lazaridis, 1986).<br />
6.1. A Note on Integration<br />
It seems to be a common belief that differentiating a variable say n times,<br />
we will always get a stationary series (Harris, 1995, p. 16). Hence, the<br />
initial series is said to be I(n). But, this is not necessarily the case, as it is<br />
verified that if we take into consideration the exports of goods and services<br />
for Greece, as presented in Table 1. Besides if n is large enough, is it of<br />
any good to obtain such a stationary series when economic variables are<br />
considered?<br />
Table 1. The series {x i }.<br />
Exports of goods and services in bil. Eur. Greece 1956–2000.<br />
X<br />
2.4944974E-02 2.9053558E-02 2.9347029E-02 2.8760089E-02 2.8173149E-02<br />
3.2575201E-02 3.5803374E-02 4.1379310E-02 4.2553190E-02 4.7248717E-02<br />
6.6030815E-02 6.7498162E-02 6.6030815E-02 7.6008804E-02 8.8041089E-02<br />
0.1000734 0.1300073 0.2022010 0.2664710 0.3325018<br />
0.4258254 0.4763023 0.5998532 0.7325019 1.049743<br />
1.239618 1.388114 1.788701 2.419956 2.868965<br />
3.618782 4.510052 4.978430 5.818929 6.485106<br />
7.690976 9.316215 9.847396 11.45708 14.08716<br />
15.39428 18.87601 20.90506 22.56992 25.51577<br />
Source: International Financial Statistics. (International Monetary Fund, Washington DC)
A Novel Method of Estimation Under Co-Integration 35<br />
The series { ∆xi}, { ∆ 2 x i }, and { ∆<br />
3 x i }<br />
Figure 2. Differences of the NI series {x i }.<br />
Using the series {x i } as shown in Table 1, we can easily verify from the<br />
corresponding graph, that even for n = 5, the series { 5 x i } is not stationary.<br />
In Fig. 2, the series {x i }, { 2 x i }, and { 3 x i } are presented, for a better<br />
understanding of this situation. We will call the variables, which belong to<br />
this category near the integrated (NI) (Banerjee et al., 1993, p. 95) series.<br />
To trace that a given series is NI, we may run the regression:<br />
x i = β 1 + β 2 x i−1 + β 3 t i +<br />
q∑<br />
β j+3 x i−j + u i (24)<br />
which is a re-parametrized AR(q) with constant, where we added a trend<br />
too. As shown in Eq. (17), the value of q is set such that the noises u i to<br />
be white. A comparatively simple way for this verification is to compute<br />
the residuals û i and to consider the corresponding Ljung-Box Q statistics<br />
and particularly their p-values, which should be much greater than<br />
0.1, to say that no autocorrelation (AC) is present. On the other hand, if<br />
we trace heteroscedasticity, this is a strong indication that we have a NI<br />
series. For the data presented in Table 1, we found that q = 2. The corresponding<br />
Q statistics (Column 4) together with p-values are presented<br />
in Table 2.<br />
We see that for all k (column 1), the corresponding p-values (column 5)<br />
are greater than 0.1. But, if we look at the residuals graph (Fig. 3), we can<br />
verify the presence of heteroscedasticity.<br />
j=1
36 Alexis Lazaridis<br />
Table 2.<br />
Residuals: Autocorrelations, PAC, Q statistics and p-values.<br />
Autocorrelation (AC), partial autocor coefficients (PAC) Q statistics and p values (prob).<br />
Values AC PAC Q Stat(L-B) p value Q Stat(B-P) p value<br />
of k<br />
1 0.040101 0.040101 0.072482 0.787756 0.067540 0.794952<br />
2 −0.061329 −0.063039 0.246254 0.884151 0.225515 0.893367<br />
3 −0.008189 −0.003064 0.249432 0.969240 0.228331 0.972891<br />
4 −0.285494 −0.290504 4.213238 0.377916 3.651618 0.455202<br />
5 0.042856 0.072784 4.304969 0.506394 3.728756 0.589090<br />
6 0.099724 0.058378 4.815469 0.567689 4.146438 0.656867<br />
7 −0.151904 −0.167680 6.033816 0.535806 5.115577 0.645861<br />
8 −0.007846 −0.069642 6.037162 0.643069 5.118163 0.744875<br />
9 0.001750 0.023203 6.037333 0.736176 5.118291 0.823877<br />
10 0.109773 0.165620 6.733228 0.750367 5.624397 0.845771<br />
11 0.134538 0.022368 7.812250 0.730017 6.384616 0.846509<br />
12 −0.029101 −0.048363 7.864417 0.795634 6.420185 0.893438<br />
13 −0.074997 −0.028063 8.222832 0.828786 6.656414 0.918979<br />
14 −0.084284 −0.017969 8.691681 0.850280 6.954772 0.936441<br />
15 −0.113984 −0.104484 9.580937 0.845240 7.500452 0.942248<br />
16 −0.081662 −0.157803 10.05493 0.863744 7.780540 0.955133<br />
17 −0.001015 −0.005026 10.05501 0.901288 7.780582 0.971016<br />
18 −0.053178 −0.056574 10.27276 0.922639 7.899356 0.980097<br />
Figure 3. The residuals from Eq. (24) (q = 2).
A Novel Method of Estimation Under Co-Integration 37<br />
Another easy and practical way to trace heteroscedasticity, is to select<br />
the explanatory variable, which yields the smallest p-value for the corresponding<br />
Spearman’s correlation coefficient (r s ). Note that in models like<br />
the one we have seen in Eq. (24), such a variable is the trend, resulting<br />
in this case to: r s = 0.3739,t = 2.394,p = 0.0219. This means that<br />
for a significance level α>0.022, heteroscedasticity is present. Hence,<br />
we have an initially NI series {x i }, which has to be weighed somehow in<br />
order to be transformed to a difference stationary series (DSS). In some<br />
cases, this transformation can be achieved if the initial series is expressed<br />
in real terms, or as percentage of growth. In usual applications, the (natural)<br />
logs [i.e., ln(x i )] are considered, given that {x i /x i } > {ln(x i )} ><br />
{x i /x i−1 }.<br />
7. Case Study 3: How to Obtain Reliable Results Regarding<br />
the Order of Integration?<br />
It is a common practice to apply DF/ADF for determining the order of integration<br />
for a given series. However, the analytical step-by-step application<br />
of this test (Enders, 1995, p. 257; Holden and Perman, 1994, pp. 64–65),<br />
leaves the impression that the cure is worse than the disease itself. On the<br />
other hand, by simply adopting the results produced by some commercial<br />
computer programs, which is the usual practice in many applications,<br />
we may reach some faulty conclusions. It should be noted at this point that<br />
according to the results obtained by a well-known commercial package, the<br />
series {x i }, as shown in Table 1, should be considered as I(2), for α = 0.05,<br />
which is not the case.<br />
Here, we present a comparatively simple procedure to determine the<br />
order of integration of a DSS, by inspecting a model analogous to Eq. (24).<br />
Assuming that {z i } is DSS, we proceed as follows:<br />
• Start with k = 1 and estimate the regression:<br />
k z i = β 1 + β 2 k−1 z i−1 + β 3 t i +<br />
q∑<br />
β j+3 k z i−j + u i . (25)<br />
Be sure that the noises in Eq. (25) are white, by applying the relevant<br />
tests as described above. This way, we set the lag-length q. Note that if<br />
k = 1, its value is omitted from Eq. (25) and that 0 z i−1 = z i−1 .<br />
j=1
38 Alexis Lazaridis<br />
Table 3.<br />
Critical values for the DF F-test.<br />
Sample size α = 0.01 α = 0.05 α = 0.10<br />
25 10.61 7.24 5.91<br />
50 9.31 6.73 5.61<br />
100 8.73 6.49 5.47<br />
250 8.43 6.34 5.39<br />
500 8.34 6.30 5.36<br />
>500 8.27 6.25 5.34<br />
• Proceed to test the hypothesis:<br />
H 0 : β 2 = β 3 = 0. (26)<br />
To compare the computed F-statistic, we have to consider, in this case,<br />
the Dickey-Fuller F-distribution. The critical values (Dickey and Fuller,<br />
1981) are reproduced in Table 3, to facilitate the presentation.<br />
Note that in order to reject Eq. (26), F-statistic should be much greater<br />
than the corresponding critical values, as shown in Table 3. If we reject<br />
Eq. (26), this means that the series {z i } is I(k − 1). If the null hypothesis<br />
is not rejected, then increase k by 1(k = k + 1) and repeat the same<br />
procedure, i.e., starting from estimating Eq. (25), testing at each stage so<br />
that the model disturbances are white noises.<br />
• In case of known structural break(s) in the series, a dummy d i should be<br />
added in Eq. (25), such that:<br />
{ 0 if i ≤ r<br />
d i =<br />
w i if i>r<br />
where w i = 1(∀ i) for the shift in mean, or w i = t i for the shift in<br />
trend. It is re-called that r denotes the date of the break or shift. With<br />
this specification, the results obtained are alike the ones that Perron et al.<br />
(1992a; b) procedure yields.<br />
We applied the proposed technique to find the order of integration of the<br />
variable m i as shown in Eq. (21), which is the log of nominal M1 (money<br />
supply). Harris (1995, p. 38) after applying DF/ADF test reports that “...<br />
and the interest rate as I(1) and prices as (I2). The nominal money supply<br />
might be either, given its path over time....”. Since the latter variable is m i ,<br />
it is clear that, with this particular test, it was not possible to get a precise<br />
answer as to whether {m i } is either I(1) or I(2).
A Novel Method of Estimation Under Co-Integration 39<br />
To show the details, underline the first stage of this procedure; we<br />
present some estimation results with different values of q.<br />
For q = 3<br />
Most of the p-values in the 5th column of the table, which is analogous<br />
to Table 2, are equal to zero.<br />
Obviously, this is not the proper lag-length.<br />
For q = 4<br />
All p-values are greater than 0.1, as shown in Table 2. r s =−0.193,<br />
p = 0.05652. F(2, 94) = 4.2524.<br />
This lag-length is acceptable.<br />
For q = 5<br />
All p-values are greater than 0.1, as shown in Table 2. r s =<br />
−0.2094,p= 0.0395.F(2, 92) = 4.2278.<br />
This lag-length is questionable, since for α>0.04, we face the problem<br />
of heteroscedasticity.<br />
q = 6<br />
All p-values are greater than 0.1, as we have seen in Table 2. r s =<br />
−0.166,p= 0.1038.F(2, 90) = 5.937.<br />
This lag-length is quite acceptable.<br />
q = 7<br />
All p-values are greater than 0.1, as shown in Table 2. r s =−0.1928,p=<br />
0.0608.F(2, 88) = 4.768.<br />
This lag-length is also acceptable, but it is inferior when compared to<br />
the previous one.<br />
From these analytical results, it can be easily verified that the proper<br />
lag-length is 6 (q = 6). It may also be worthy to mention that according<br />
to the normality test (Jarque-Bera), we should accept the null. The value<br />
of F = 5.937 is less than the corresponding critical values as shown in<br />
Table 3, for 100 observations and α = (0.01, 0.05). Hence, we accept<br />
Eq. (26) and increase the value of k by one. The estimation results, after<br />
properly selecting the value of q(q = 5), are:<br />
2 m i = 0.0047 − 0.894892 m i−1 + 0.00033 t i<br />
(0.0046) (0.21784) (0.000108)<br />
+<br />
5∑<br />
ˆβ j+3 2 m i−j +û i . (27)<br />
j=1<br />
Regarding the residuals, all p-values in the 5th column of the table,<br />
which is analogous to Table 2, are greater than 0.1. Further, we have:<br />
r s =−0.1556,p= 0.127. Jarque-Bera = 0.8(p = 0.67).F(2, 91) = 8.44.
40 Alexis Lazaridis<br />
Figure 4. The series {m i } and { 2 m i }.<br />
Since F
A Novel Method of Estimation Under Co-Integration 41<br />
Due to the properties of SVD, we can easily and directly include in the<br />
co-integrating vectors, as many deterministic components as we want to.<br />
Besides, we found that the errors corresponding to the finally selected cointegration<br />
vector, are characterized by the minimum variance property. It<br />
should be noted that the procedure proposed here is not mentioned in the<br />
relevant literature.<br />
Further, we analyzed the stages of a relevant technique to obtain reliable<br />
results regarding the order of integration of a given series provided that<br />
it is DSS. We also demonstrated the properties of the so-called near integrated<br />
series, which can be traced by applying this technique. It should be<br />
emphasized that with this kind of series (NI), we show that the conventional<br />
methods for unit root test may produce faulty results.<br />
References<br />
Banerjee, A, JJ Dolado, JW Galbraith and DF Hendry (1993). Co-integration, Error Correction,<br />
and the Econometric Analysis of Non-Stationary Data. New York: Oxford<br />
University Press.<br />
Dickey, DA and WA Fuller (1981). Likelihood ratio statistics for autoregressive time series<br />
with a unit root. Econometrica 49, 1057–1072.<br />
Dickey, DA, DW Jansen and DL Thornton (1994). A primer on cointegration with an application<br />
to money and income. In Cointegration for the Applied Economist, BB Rao<br />
(ed.), UK: St. Martin’s Press, 9–45.<br />
Enders, W (1995). Applied Econometrics Time Series. New York: John Wiley.<br />
Harris, R (1995). Using Cointegration Analysis in Econometric Modelling. London: Prentice<br />
Hall.<br />
Holden, D and R Perman (1994). Unit roots and cointegration for the economist. In Cointegration<br />
for the Applied Economist, BB Rao (ed.), UK: St. Martin’s Press, 47–112.<br />
Johansen, S and K Juselius (1990). Maximum likelihood estimation and inference on cointegration<br />
with application to the demand for money. Oxford Bulletin of <strong>Economic</strong>s and<br />
Statistics 52, 169–210.<br />
Johnston, J and J DiNardo (1997). Econometric <strong>Models</strong>. New York: McGraw-Hill International.<br />
Kleibergen, F and V Dijk (1994). Direct cointegration testing in error correction models.<br />
Journal of Econometrics 63, 61–103.<br />
Lazaridis, A and D Basu (1983). Stochastic optimal control by pseudo-inverse. Review of<br />
<strong>Economic</strong>s and Statistics 65(2), 347–350.<br />
Lazaridis, A (1986). A note regarding the problem of perfect multicollinearity. Quantity and<br />
Quality 120, 297–306.<br />
McKinnon, J (1991). Critical values for co-integration tests. In Long-Run <strong>Economic</strong> Relationships,<br />
RF Engle and CWJ Granger (eds.), New York: Oxford University Press.<br />
Perron, P and T Vogelsang (1992a). Nonstationarity and level shifts with an application<br />
to purchasing power parity. Journal of Business and <strong>Economic</strong> Statistics 10,<br />
301–320.
42 Alexis Lazaridis<br />
Perron, P and T Vogelsang (1992b). Testing for a unit root in a time series with a changing<br />
mean: Corrections and extensions. Journal of Business and <strong>Economic</strong> Statistics 10,<br />
467–470.<br />
Sims, CA (1980). Macroeconomics and reality. Econometrica 48, 1–48.
Topic 2<br />
Time Varying Responses of Output to Monetary<br />
and Fiscal Policy<br />
Dipak R. Basu<br />
Nagasaki University, Nagasaki, Japan<br />
Alexis Lazaridis<br />
Aristotle University of Thessalonikki, Greece<br />
1. Introduction<br />
There are large volumes of literatures on the effectiveness of monetary and<br />
fiscal policies and on lags in the effects of monetary policies (Friedman and<br />
Schwartz, 1963; Hamburger, 1971; Meyer, 1967; Tobin, 1970). Although<br />
the lengths of lags are important in determining the role of monetary-fiscal<br />
policies, the relationship between money and income vary over time due<br />
to changes in the lag structure. Mayer (1967), Poole (1975) and Warburton<br />
(l971) attempted to analyze the empirical variability of the lag structure<br />
by analyzing the turning points in general business activity. Sargent and<br />
Wallace (1973) as well as Lucas (1972) have drawn attention to the role of<br />
time-dependant response coefficients to changes in stabilization policies.<br />
Cargill and Meyer (1977; 1978) have estimated the time-varying relationship<br />
between national income and monetary-fiscal policies. The results<br />
then indicated existences of time variations and exclusions of time variations<br />
of the coefficients, lead to exclusions of prior information inherent in<br />
the models from the estimation process. Blanchard and Perotti (2002) as<br />
well as Smets and Wouters (2003) have obtained similar characteristics of<br />
the monetary and fiscal policy.<br />
The existence of time dependency of effects of monetary fiscal policies,<br />
reflects considerable doubts on the policy prescription based on constant<br />
coefficient estimates. However, more reasonable results can be obtained if<br />
we try to estimate dynamics and movements of these relationships over<br />
43
44 Dipak R. Basu and Alexis Lazaridis<br />
time. For this purpose, adaptive control systems with time-varying parameters<br />
(Astrom and Wittenmark, 1995; Basu and Lazaridis, 1986; Brannas<br />
and Westlund, 1980; Nicolao, 1992; Radenkovic and Michel, 1992) provide<br />
a fruitful approach. There are alternative estimation methods, e.g., rational<br />
expectation modeling, VAR approaches etc., to estimate time-varying systems.<br />
These methods, however, are mainly descriptive, i.e., cannot have<br />
immediate application in policy planning. The purpose of this paper is to<br />
explore this possibility in terms of a planning model where adaptive control<br />
rules will be implemented so that we can derive time-varying reduced form<br />
coefficients from the original structural model to demonstrate the dynamics<br />
of monetary-fiscal policies.<br />
The model follows the basic theoretical ideas of monetary approach to<br />
balance of payments and structural adjustments (Berdell, 1995; Humphrey,<br />
1981; 1993; Khan and Montiel, 1989). In Section 2, the method of adaptive<br />
optimization and the updating method of the time-varying model are analyzed.<br />
In Section 3, the model and its estimates are described. This includes<br />
some necessary adjustments for a developing country. In Section 4, the<br />
results of the time-varying impacts of monetary-fiscal policies are analyzed.<br />
2. The Method of Adaptive <strong>Optimization</strong><br />
Suppose a dynamic econometric model can be converted to an equivalent<br />
first order dynamic system of the form<br />
˜x i = Øx i−1 + ˜Cũ i + ˜D˜z i +ẽ i (1)<br />
where ˜x i is the vector of endogenous variables, ũ i is the vector of control<br />
variables, ˜z i is the vector of exogenous variables, and ẽ i is the vector<br />
of noises, which are assumed to be white Gaussian and Ã, ˜C, and ˜D are<br />
coefficient matrices of proper dimensions. It should be noted that a certain<br />
element of ˜z i is 1 and corresponds to the constant terms. The parameters of<br />
the above system are assumed to be random.<br />
Shifting to period i + 1, we can write<br />
˜x i+1 = Øx i + ˜Cũ i+1 + ˜D˜z i+1 +ẽ i+1 . (2)<br />
Now, we define the following augmented vectors and matrices.<br />
[ ] [ ] [ ]<br />
˜xi<br />
˜xi+1<br />
ẽi+1<br />
x i = , x<br />
ũ i+1 = , e<br />
i ũ i+1 = ,<br />
i+1 0<br />
A =<br />
[ ]<br />
à 0<br />
, C =<br />
0 0<br />
[ ] ˜C<br />
, D =<br />
I<br />
[ ] ˜D<br />
.<br />
0
Time Varying Responses 45<br />
Hence (2) can be written as<br />
x i+1 = Ax i + Cũ i+1 + D˜z i+1 + e i+1 (3)<br />
Using the linear advance operator L, such that L k y i = y i+k and defining<br />
the vectors u, z and ε from<br />
then Eq. (3) can take the form<br />
u i = Lũ i<br />
z i = L˜z i<br />
ε i = Le i<br />
x i+1 = Ax i + Cu i + Dz i + ε i (4)<br />
which is a typical linear control system.<br />
We can formulate an optimal control problem of the general form<br />
T −1<br />
min J = 1 2 ‖x T −ˆx T ‖ 2 Q T<br />
+ 1 ∑<br />
‖x i −ˆx i ‖ 2 Q<br />
2<br />
i<br />
(5)<br />
subject to the system transition equation shown in Eq. (4).<br />
It is noted that T indicates the terminal time of the control period, {Q}<br />
is the sequence of weighing matrices and ˆx i (i = 1, 2,...,T)is the desired<br />
state and control trajectory, according to our formulation.<br />
The solution to this problem can be obtained according to the minimization<br />
principle by solving the Ricatti-type equations (Astrom and<br />
Wittenmark, 1995).<br />
K T = Q T (6)<br />
i =−(E i C ′ K i+1 C) −1 (E i C ′ K i+1 A) (7)<br />
K i = E i A ′ K i+1 A + ′ i (E iC ′ K i−1 A) + Q i (8)<br />
h T =−Q T ˆx T (9)<br />
h i = i (E i C ′ K i+1 D)z i + i (E i C ′ )h i+1<br />
+ (E i A ′ K i+1 D)z i + (E i A ′ )h i+1 − Q i ˆx i (10)<br />
g i =−(E i C ′ K i+1 C) −1 [(E i C ′ K i+1 D)z i + (E i C ′ )h i+1 ] (11)<br />
xi ∗ = [E i A + (E i C) i ] xi ∗ + (E iC) g i + (E i D)z i (12)<br />
u ∗ i = i xi ∗ + g i (13)<br />
i=1
46 Dipak R. Basu and Alexis Lazaridis<br />
where u ∗ i (i = 0, 1,...,T − 1), the optimal control sequence and x∗ i+1 , the<br />
corresponding state trajectory, which constitutes the solution to the stated<br />
optimal control problem.<br />
In the above equations, i is the matrix of feedback coefficients and<br />
g i is the vector of intercepts. The notation E i denotes the conditional<br />
expectations, given all information up to the period i. Expressions like<br />
E i C ′ K i+1 C, E i C ′ K i+1 A, E i C ′ K i+1 D are evaluated, taking into account<br />
the reduced form coefficients of the econometric model and their covariance<br />
matrix, which are to be updated continuously, along with the implementation<br />
of the control rules. These rules should be re-adjusted according<br />
to “passive-learning” methods, where the joint densities of matrices A, C,<br />
and D are assumed to remain constant over the control period.<br />
2.1. Updating Method of Reduced-Form Coefficients and Their<br />
Co-Variance Matrices<br />
Suppose we have a simultaneous-equation system of the form<br />
XB ′ + UƔ ′ = R (14)<br />
where X is the matrix of endogenous variable defined on E N × E n and<br />
B is the matrix of structural coefficients, which refer to the endogenous<br />
variables and is defined on E n ×E n . U is the matrix of explanatory variables<br />
defined on E N ×E g and Ɣ is the matrix of the structural coefficients, which<br />
refer to the explanatory variables, defined on E N × E g . R is the matrix of<br />
noises defined on E N ×E n . The reduced form coefficients matrix is then<br />
defined from:<br />
=−B −1 Ɣ.<br />
(14a)<br />
Goldberger et al. (1961) have shown that the asymptotic co-variance<br />
matrix, say of the vector ˆπ, which consists of the g column of matrix ˆ<br />
can be approximated by<br />
˜ =<br />
[[ ˆ<br />
I g<br />
]<br />
⊗ ( ˆB ′ ) −1 ] ′<br />
F<br />
[[ ˆ<br />
I g<br />
]<br />
⊗ ( ˆB ′ ) −1 ]<br />
(15)<br />
where ⊗ denotes the Kroneker product, ˆ and ˆB are the estimated coefficients<br />
by standard econometric techniques and F denotes the asymptotic<br />
co-variance matrix of the n + g columns of ( ˆB ˆƔ), which assumed to be<br />
consistent and asymptotically unbiased estimate of (BƔ).
Time Varying Responses 47<br />
Combining Eqs. (14) and (14a) we have<br />
BX ′ =−ƔU ′ + R ′ ⇒ X ′ =−B −1 ƔU ′ + B −1 R ′<br />
⇒ X ′ = U ′ + W ′ (16)<br />
where W ′ = B −1 R ′<br />
Denoting the ith column of matrix X ′ by x i and the ith column of matrix<br />
W ′ by w i , we can write<br />
⎡<br />
⎤<br />
u 1i 0 ··· 0 u 2i 0 ··· 0 u gi 0 ··· 0<br />
0 u 1i ··· 0 0 u 2i ··· 0 0 u gi ··· 0<br />
x i =<br />
· · · · · · · · ·<br />
⎢ · · · · · · · · ·<br />
⎥<br />
⎣ · · · · · · · · · ⎦<br />
0 0 ··· u 1i 0 0 ··· u 2i ··· 0 ... u gi<br />
π + w i<br />
(17)<br />
where u ij is the element of the jth column and ith row of matrix U. The<br />
vector π ∈ E ng , as mentioned earlier, consists of the g column of matrix .<br />
Equation (17) can be written in a compact form, as<br />
x i = H i π + w i , i = 1, 2,...,N (17a)<br />
where x i ∈ E n ,w i ∈ E n and the observation matrix H i is defined on<br />
E n × E ng .<br />
In a time-invariant econometric model, the coefficients vector π is<br />
assumed random with constant expectation overtime, so that<br />
π i+1 = π i , for all i. (18)<br />
In a time-varying and stochastic model we can have<br />
π i+1 = π i + ε i<br />
(18a)<br />
where ε i ∈ E ng is the noise.<br />
Based on these, we can re-write Eq. (17a) as<br />
x i+1 = H i+1 π i+1 + w i+1 i = 0, 1,...,N − 1. (19)<br />
We make the following assumptions.<br />
(a) The vector x i+1 and matrix H i+1 can he measured exactly for all i.<br />
(b) The noises ε i and w i+1 are independent discrete white noises with<br />
known statistics, i.e.,<br />
E(ε i ) = 0; E(w i+1 ) = 0<br />
E(ε i w ′ i+1 ) = 0
48 Dipak R. Basu and Alexis Lazaridis<br />
E(c i ε ′ i ) = Q 1δ ij ,<br />
E(w i w ′ i ) = Q 2δ ij .<br />
where δ ij is the Kronecker delta, and<br />
The above co-variance matrices, assumed to be positive definite.<br />
(c) The state vector is normally distributed with a finite co-variance matrix.<br />
(d) Regarding Eqs. (18a) and (19), the Jacobians of the transformation of<br />
ε i into π i+1 and of w i+1 into x i+1 are unities. Hence, the corresponding<br />
conditional probability densities are:<br />
p(π i+1 |π i ) = p(ε i )<br />
p(x i+1 |π i+1 ) = p(w i+1 ).<br />
Under the above assumptions and given Eqs. (18a) and (19), the problem<br />
set is to evaluate<br />
and<br />
E(π i+1 |x i+1 ) = π ∗ i+1<br />
cov (π i+1 |x i+1 ) = S i+1<br />
(the error co-variance matrix)<br />
where x i+1 = x 1 ,x 2 ,x 3 ,...,x i+1 .<br />
The solution to this problem (Basu and Lazaridis, 1986; Lazaridis,<br />
1980) is given by the following set of recursive equations, as it is briefly<br />
shown in Appendix A.<br />
π ∗ i+1 = π∗ i + K i+1(x i+1 − H i+1 π ∗ i ) (20)<br />
K i+1 = S i+1 H i+1 ′ Q−1 2<br />
(21)<br />
Si+1 −1 i+1 + H i+1 ′ Q−1 2 H i+1 (22)<br />
Pi+1 −1 1 + S i ) −1 . (23)<br />
The recursive process is initiated by regarding K 0 and H 0 as null matrices<br />
and computing π ∗ 0 and S 0 from<br />
π0 ∗ =ˆπ i.e., the reduced form coefficients (columns of matrix ˆ)<br />
S 0 = P 0 = ˜.<br />
The reduced form coefficients, along with their co-variance matrices,<br />
can be updated by this recursive process and at each stage the set of Riccati
Time Varying Responses 49<br />
equations should be updated accordingly so that adaptive control rules may<br />
be derived.<br />
Once we estimate the model (described in the next section), using the<br />
full information maximum likelihood (FIML) method, we can obtain both<br />
the structural model and the probability density functions along with all<br />
associated matrices mentioned above. We first convert the structural econometric<br />
model to a state-variable form according to Eq. (1). Once we specify<br />
the targets for the state and control variables, the objective function is to be<br />
minimized, the weights attached to each state and control variables, then<br />
we can calculate the results of the optimization process for the entire period<br />
using Eqs. (6)–(13). Thereafter, we can update all probability density functions<br />
and all other associated matrices using Eqs. (20)–(23). These will<br />
effectively update the coefficients of the model in its state-variable form.<br />
We can repeat the optimization process over and over as we update the<br />
model, its associated matrices, probability density functions and use these<br />
as new information. When the process will converge, i.e., the difference<br />
between the old and new estimates are insignificant according to some preassigned<br />
criteria, we can obtain the final updated coefficients of the model<br />
along with the results of the optimization process. The dynamics of the<br />
time-varying coefficients (in terms of response-multipliers) are described<br />
later in Section 4. This mechanism is a great improvement compared to the<br />
fixed-coefficient model as we can incorporate in our calculation, changing<br />
information sets derived from the implementation of the optimization<br />
process.<br />
3. Monetary-Fiscal Policy Model<br />
We describe below a monetary-fiscal policy model, which was estimated<br />
with data from the Indian economy. The model accepts the definition of<br />
balance of payments and money stock according to monetary approach<br />
of balance of payments (Khan, 1976; Khan and Montiel, 1989; 1990).<br />
The decision to choose this particular model is influenced by the fact that<br />
it is commonly known as the fund-bank adjustment policy model, used<br />
extensively by the World Bank and the International Monetary Fund (IMF).<br />
In the version adopted for this paper, there is no explicit investment<br />
or consumption function, because the domestic absorption can reflect the<br />
combined response of both private and public investment and consumption,<br />
to the planned target for national income set by the planning authority<br />
and to various market forces reflected in the money market interest
50 Dipak R. Basu and Alexis Lazaridis<br />
rate and exchange rate. The “New Cambridge” model of the UK economy<br />
(Cripps and Godley, 1976; Godley and Lavoie, 2002; 2007) has postulated<br />
a similar combined consumption-investment function for the UK<br />
as well.<br />
The model was estimated by applying FIML method, using the data<br />
from the Indian economy for the period 1951–1996. The estimated parameters<br />
were then used as the initial starting point for the stochastic control<br />
model. The estimated econometric model can be transformed into a state<br />
variable form (Basu and Lazaridis, 1986), in order to formulate the optimal<br />
control problem, i.e., to minimize the quadratic objective function, subject<br />
to Eq. (4), which is the system transition equation. As already mentioned,<br />
the recursive estimation process of the time-varying response-multipliers<br />
is presented in brief in Appendix A.<br />
3.1. Absorption Function and National Income<br />
Domestic absorption reflects the behavior of both the private and public<br />
sector regarding consumption and investment too. Domestic real adsorption<br />
is influenced by real national income, market interest rate and foreign<br />
exchange rate. We assume a linear relationship.<br />
(A/P) t = a 0 + a 1 (Y/P) t − a 2 (IR) t − a 3 EXR t ,<br />
t = time period.<br />
The relation between the national income and adsorption can be defined<br />
as follows:<br />
Y t = A t + TY t + R t − G t + GBS t − LR t<br />
where A is the value of domestic absorption, P is the price level, Y is the<br />
national income, IR is the market interest rates, EXR is the exchange rate,<br />
TY is the government tax revenue, G is the public consumption, GBS is the<br />
government bond sales, LR is the net lending by the central government to<br />
the states (which is not part of the planned public expenditure) and R is<br />
the changes in the foreign exchange reserve reflecting the behavior of the<br />
foreign trade sector.<br />
The government budget deficit (BD t ) is defined by the following<br />
equation<br />
BD t = (G t + LR t + PF t ) − (TY t + GBS t + AF t + BF t )<br />
where PF is the foreign payments due to existing foreign debts, which<br />
may include both amortization and interest payments, AF is the foreign
Time Varying Responses 51<br />
assistance which is an insignificant feature, FB is the total foreign borrowing,<br />
assuming only the government can borrow from foreign sources.<br />
We assume AF and LR as exogenous, whereas FB, G, and TY as policy<br />
instruments. However, PF depends on the level of existing foreign debt and<br />
the world interest rate, although a sizable part of the foreign borrowing can<br />
be at a concessional rate<br />
PF t = a 4 + a 5<br />
t∑<br />
r=−20<br />
( ) WIR<br />
FB r + a 6 .<br />
EXR t<br />
Government bond sales (GBS) depends on its attractiveness reflected on<br />
the interest rate (IR), on the ability of the domestic economy to absorb (A)<br />
on the requirements of the government (G) and on the alternative sources of<br />
finances reflected on the tax revenue (TY) and on government’s borrowing<br />
from the central bank i.e., the net domestic asset (NDA) creation by the<br />
central bank.<br />
GBS t = a 7 + a 8 A t + a 9 IR t + a 10 G t − a 11 TY t − a 12 NDA t<br />
3.2. Monetary Sector<br />
We assume the flow equilibrium in the money market, i.e.,<br />
MD t = M t = MS t<br />
where MD is the money demand and MS is the money supply. The stock<br />
of money supply depends on the stock of high powered money and the<br />
money-multiplier, as follows:<br />
MS t = [(1 + CD)/(CD + RR)] t (R + NDA) t .<br />
(R + NDA) reflect the stock of high-powered money and the expression<br />
within the square bracket is the money-multiplier, which depends on credit<br />
to deposit ratio of the commercial banking sector (CD) and the reserve to<br />
deposit liabilities in the commercial banking sector (RR). Whereas NDA<br />
is an instrument, R depends on the foreign trade sector. However, the<br />
government can influence CD and RR to control the money supply. RR,<br />
which is the actual reserve ratio, depends on the demand for loans created by<br />
private sectors and the commercial bank’s willingness to lend. We assume<br />
that the desired reserve ratio RR ∗ t is a function of national income and
52 Dipak R. Basu and Alexis Lazaridis<br />
market interest rate i.e.,<br />
RR ∗ t = a 13 + a 14 Y t + a 15 IR ∗ t .<br />
The commercial banks may adjust their actual reserve ratio to the<br />
desired reserve ratio with a lag.<br />
Thus, we can write<br />
RR t = α(RR ∗ t − RR t−1 )<br />
where 0
Time Varying Responses 53<br />
The import cost in turn depends on the exchange rate (EXR) and the world<br />
price of imported goods (WPM). We assume the desired price level (P ∗ )is<br />
represented by the following equation:<br />
P ∗ t = a 27 − a 28 (A t ) + a 29 (IMC t ).<br />
The desired price level reflects private sector’s reaction to their expected<br />
domestic absorption of the expected import cost. Suppose the actual price<br />
will move according to the difference between the desired price in period t<br />
and the actual price level in the previous period<br />
Pt = β(P ∗ t − P t−1 );<br />
0
54 Dipak R. Basu and Alexis Lazaridis<br />
3.5. Stability of the Model<br />
Characteristic roots (real) of the system transition matrix<br />
− 0.0000008<br />
− 0.0000008<br />
− 0.1213610<br />
0.2442519<br />
0.3087129<br />
0.5123629<br />
0.7824260<br />
0.0000001<br />
0.0000001<br />
All the roots are real and they are less than unity, so the system is<br />
stable. [In the equivalent control system, the rank of the controllability<br />
matrix is the same as the dimension of the reduced state vector so that the<br />
system is controllable and observable, i.e., the system parameters can be<br />
identified.]<br />
We can, however, transform our model to the following form:<br />
Y t = ρY t−1 + e t , t = 1, 2,...,n (24)<br />
where ρ is a real number and (e t ) is a sequence of normally distributed<br />
random variables with mean zero and variance σt 2 . Box and Pierce (1970)<br />
suggested the following test statistic<br />
where<br />
Q m = n<br />
m∑<br />
rk 2 (25)<br />
k=1<br />
/(<br />
n∑<br />
n∑<br />
r k = ê t ê t−k<br />
t=k+1 t=1<br />
ê 2 t<br />
)<br />
(26)<br />
n = the number of observations, m = n − k, where k = the number of<br />
parameters estimated, and ê t are the residuals from the fitted model.<br />
If (Y t ) satisfies the system, then under the null hypothesis, Q m is<br />
distributed as a chi-squared random variable with m degrees of freedom.<br />
The null hypothesis is that ρ = 1 where ê t = Y t − Y t−1 and thus k = 0.
Time Varying Responses 55<br />
The estimated value of Q m is 3.85, where the null hypothesis is rejected<br />
at 0.05 significance level. So, we accept the alternative hypothesis that if<br />
ρ
56 Dipak R. Basu and Alexis Lazaridis<br />
Table 4. Response-multiplier period 3.<br />
Endogenous Y GBS P BD CD MS IR<br />
EXR −0.02199 −0.00069 0.05269 0.00133 −0.00042 −0.00573 −0.00062<br />
G 0.03001 0.00499 0.16536 0.02924 0.05923 0.04217 0.00570<br />
TY −0.02470 −0.01149 −0.16281 0.02472 −0.07031 −0.06325 −0.02220<br />
CI 0.00215 0.04676 −0.00252 −0.00399 −0.00465 −0.00671 0.06228<br />
Table 5. Response-multiplier period 5.<br />
Endogenous Y GBS P BD CD MS IR<br />
EXR −0.02133 −0.00088 0.05190 0.00152 −0.00201 −0.00595 −0.00089<br />
G 0.02058 0.00558 0.18078 0.01160 0.50890 0.07248 0.00530<br />
TY −0.01676 −0.00410 −0.16655 0.00695 −0.55026 −0.07779 −0.01013<br />
CI 0.00268 0.04465 −0.00267 −0.00403 −0.01967 −0.00726 0.05953<br />
Table 6. Response-multiplier period 7.<br />
Endogenous Y GBS P BD CD MS IR<br />
EXR −0.02067 −0.00116 0.04987 0.00135 −0.00104 −0.00558 −0.00129<br />
G 0.00733 0.01033 0.16455 0.00048 0.57357 0.10760 0.01206<br />
TY −0.00402 −0.00345 −0.14832 −0.01434 −0.58067 −0.10904 −0.00869<br />
CI 0.00200 0.04174 −0.00236 −0.00293 −0.02186 −0.00551 0.05566<br />
Table 7.<br />
Response of output to monetary and fiscal policies.<br />
Periods 1 2 3 5 7<br />
dY/dCI −0.00226 −0.00273 −0.00215 −0.00268 −0.00200<br />
dY/dEXR −0.02254 −0.02209 −0.02188 −0.02133 −0.02067<br />
dP/dG 0.14849 0.15603 0.16536 0.18078 0.16455<br />
dY/dG 0.03572 0.03203 0.03001 0.02058 0.00733<br />
dP/dTY −0.15549 −0.15688 −0.16281 −0.16655 −0.14832<br />
dY/dTY −0.03590 −0.02691 −0.02470 −0.01676 −0.00402<br />
dGBS/dG 0.00403 0.00568 0.00499 0.00558 0.01033
Time Varying Responses 57<br />
4. Policy Analysis<br />
The dynamics of the monetary and fiscal policy are analyzed in terms of<br />
the analysis of the dynamic response-multipliers and their relationship with<br />
the monetary-fiscal policy regimes. The response-multiplier of the system<br />
will move over time within an adaptive control framework (Tables 1–6).<br />
The movements of the coefficients response of response-multipliers in a<br />
monetary policy framework are described below. It is essential that for the<br />
stability of the control system, these dynamics of the response-multiplier<br />
should be slow (Das and Cristi, 1990; Tsakalis and Ioannou, 1990).<br />
Analysis of the response-multipliers shows that devaluation would have<br />
negative effect on the national income, devaluation would also have negative<br />
effect on the government bond sales, CD, interest rate, and money<br />
demand. This is due to the fact that Indian exports may not be that elastic<br />
in response to devaluation, whereas devaluation will reduce import abilities<br />
significantly. As a result, national income and domestic activity would<br />
have negative effects, which would depress the activity of the private sector.<br />
Because of this CD will decline.<br />
Due to the downturn of the economic activity, there will be less demand<br />
for government bonds. Devaluations also can have an inflationary effect due<br />
to increased import costs. Government expenditure, on the other hand, will<br />
have positive effects on every variable. This implies that increased public<br />
expenditure will stimulate the national income and the private sector despite<br />
increased interest rate. However, the increased public expenditure will have<br />
inflationary impacts on the economy at the same time.<br />
Increased tax revenue will depress national income, as it will reduce<br />
government bond sales. Lower level of national income will reduce money<br />
supply and the private sector’s activity, which will be reflected in the reduced<br />
currency to deposit ratio and the reduced level of money demand. Reduced<br />
level of national income, in this case, will have reduced price level but budget<br />
deficit will go up as well because of the reduced demand for government<br />
bonds. Increased rate of central bank’s discount rate will depress national<br />
income and general price level. At the same time, private sector’s activity<br />
will be reduced, as reflected in the CD and money demand, due to increased<br />
rate of interest. Although government bond sales will go up, budget deficit<br />
will be increased due to reduced level of economic activity.<br />
The effect of the central bank discount rate (CI ) on interest rate varies<br />
from 0.061 in period 1 to 0.055 in period 7. This result is primarily due to<br />
the fact that the influence of G (public expenditure) on IR has increased<br />
from 0.003 in period 1 to 0.012 in period 7. If the public expenditure goes
58 Dipak R. Basu and Alexis Lazaridis<br />
up, there will be direct pressure on the central bank to provide loans to the<br />
government. The central bank can put pressure on the commercial banks to<br />
extend some loans in public sector and less to the private sector, and the CD<br />
of the commercial banks will be reduced as a result (at the same time reserve<br />
ratio will be increased). While CD goes up in response to government<br />
expenditure (G), the CI will have a similar effect on the currency to deposit<br />
ratio.<br />
Effects of the CD on prices declines rapidly, whereas the effect of the<br />
exchange rate on price level is most prominent and does not vary much. It<br />
demonstrates that although during the initial phase of planning, tight credit<br />
policy may influence prices, in the longer run structural factors, public<br />
expenditure, and import costs reflected through the exchange rate will affect<br />
price levels more significantly. The effect of the CI on the money supply<br />
(MS) will be reduced from −0.005 in period 1 to −0.007 in period 6, and<br />
it will slightly increase in period 7.<br />
The impacts of the public expenditure and tax revenues on the two most<br />
important variables, GNP (Y) and price level (P) can be analyzed as follows.<br />
The increased government expenditure (G) will increase price level and<br />
the impact will be intensified. The impact of the government expenditure<br />
on the GNP will be reduced, which is consistent with the reduced impacts<br />
of the tax revenues on the GNP over the planning period. Tax revenue will<br />
have negative impacts on the GNP as well as on the price level, and the<br />
impacts will decline over time. The negative impacts of the tax revenue on<br />
the GNP is partly explained by the negative effects of tax revenue on the<br />
currency to deposit ratios of the commercial banks, the main indicator for<br />
the private sector activities. The government expenditure also have positive<br />
impacts on the government bond sales and the impact will increase over<br />
time due to the increasing difficulties of raising taxes, which will have<br />
negative impacts on the growth prospects.<br />
The effect of the exchange rate declines gradually, whereas the effect of<br />
the interest rate is cyclical. This is due to the cyclical pattern of the central<br />
discount rate in the optimum path. It is quite obvious that the impacts of<br />
fiscal and monetary policies on output will fluctuate over time depending<br />
on the direction of change of these policies. The variations of the effects are<br />
not unsystematic, as feared by Friedman and Schwartz (1963). However,<br />
here we are analyzing an optimum path not the historical path.<br />
5. Conclusion<br />
The importance of time-varying coefficients was recognized in statistical<br />
literature long ago (Davis, 1941), where frequency domain approach of
Time Varying Responses 59<br />
the time series analysis was utilized (Mayer, 1972). Several authors since<br />
then have used varying parameter regression analysis (Cooley and Prescott,<br />
1973; Farley et al., 1975). In the above analysis, adaptive control system<br />
was used in a state-space model where the reduced form parameters can<br />
move over time.<br />
However, variations over time are slow which indicates any absence<br />
of explosive response (Das and Cristi, 1990; Tsakalis and Ioannou, 1990).<br />
Cargil and Mayer (1978) also have observed stable movements of the coefficients.<br />
Thus the role of the monetary-fiscal policies on the economy is not<br />
unsystematic, although it can vary over time.<br />
Results obtained by other researchers showed that the effects of<br />
monetary and fiscal policy change over time, and it is important to analyze<br />
these changes in order to obtain time-consistent monetary-fiscal policy (De<br />
Castro, 2006; Folster and Henrekson, 2001; Muscatelli and Tirelli, 2005).<br />
The results obtained by using adaptive control method, showed similar<br />
characteristics of the monetary-fiscal policy.<br />
The implication for the public policy is quite obvious. Time-consistent<br />
monetary-fiscal policy demands continuous revision, otherwise, the effectiveness<br />
of the policy may deteriorate and as a result, the effects of monetary<br />
and fiscal policy on major target variables of the economy may deviate from<br />
their desired level. Our approach is a systematic way forward to analyze<br />
these dynamics of monetary and fiscal policy.<br />
References<br />
Astrom, KJ and K Wittenmark (1995). Adaptive Control. Reading, MA: Addison-Wesley.<br />
Basu, D andA Lazaridis (1986).A method of stochastic optimal control by Bayesian filtering<br />
techniques. International Journal of System Sciences 17(1), 81–85.<br />
Brannas, K and A Westlund (1980). On the recursive estimation of stochastic and time<br />
varying parameters in economic system. In <strong>Optimization</strong> Techniques, K Iracki (ed.),<br />
Berlin: Springer Verlag, 265–274.<br />
Berdell, JF (1995). The prevent relevance of Hume’s open-economy monetary dynamics.<br />
<strong>Economic</strong> Journal l05(432), September, 1205–1217.<br />
Box, GEP and DA Pierce (1970). Distribution of residual autocorrelation in autoregressiveintegrated<br />
moving average time series models. Journal of the American Statistical<br />
Association 65(3), 1509–1526.<br />
Blanchard, O and R Perotti (2002). An empirical characterization of the dynamic effects of<br />
changes in government spending and taxes on output. Quarterly Journal of <strong>Economic</strong>s,<br />
117(4), November, 1329–1368.<br />
Cargil, TF and RA Meyer (1977). Intertemporal stability of the relationship between interest<br />
races and prices. Journal of Finance 32, September, 427–448.<br />
Cargil, TF and RA Mayer (1978). The time varying response of income to changes in<br />
monetary and fiscal policy. Review of <strong>Economic</strong>s and Statistics LX(1), February, 1–7.
60 Dipak R. Basu and Alexis Lazaridis<br />
Cripps, F and WAH Godley (1976). A formal analysis of the Cambridge economic policy<br />
group Model. <strong>Economic</strong>a 43, August, 335–348.<br />
Cooley, TF and EC Prescott (1973). An adaptive regression model. International <strong>Economic</strong><br />
Review l4, June, 248–256.<br />
Das, M and R Cristi (1990). Robustness of an adaptive pole placement algorithm<br />
in the presence of bounded disturbances and slow time variation of parameters.<br />
IEEE Transactions on Automatic Control 35(6), June, 762–802.<br />
Davis, HT (1941). The Analysis of <strong>Economic</strong> Time Series. Bloomington: Principia Press.<br />
De Castro, F (2006). Macroeconomic effects of fiscal policy in Spain. Applied <strong>Economic</strong>s<br />
38(8–10), May, 913–924.<br />
Farley, J, M Hinich and T McGuire (1975). Some comparison of tests for a shift in the slopes<br />
of a multivariate time series model. Journal of Econometrics 3, August, 297–318.<br />
Friedman, M and A Schwartz (1963). A monetary history of the United States, 1867–1960.<br />
Princeton: Princeton University Press.<br />
Folster, S and M Henrekson (2001). Growth effects of government expenditure and taxation<br />
in rich countries. European <strong>Economic</strong> Review 45(8), 1501–1520.<br />
Goldberger, AC, AL Nagar and HS Odeh (1961). The covariance matrices of reduced<br />
form coefficients and forecasts for a structural econometric model. Econometrica 29,<br />
556–573.<br />
Godley, W and M Lavoie (2002). Kaleckian models of growth in a coherent stock-flow<br />
monetary framework: A Kaldorian view. Journal of Post Keynesian <strong>Economic</strong>s 24(2),<br />
277–312.<br />
Godley, W and M Lavoie (2007). Fiscal policy in a stock-flow consistent (SFC) model.<br />
Journal of Post Keynesian <strong>Economic</strong>s 30(1), 79–100.<br />
Hamberger, MJ (1971). The lag in the effect of monetary policy:A survey of recent literature.<br />
Monthly Review-Federal Reserve Bank of New York, December, 289–297.<br />
Humphrey, TM (1981). Adam Smith and the monetary approach to the balance of payments.<br />
<strong>Economic</strong> Review, Federal Reserve Bank of Richmond 67.<br />
Humphrey, TM (1993). Money, Ranking and Inflation. Aldershot: Edward Elgar.<br />
Khan, MS (1976).A monetary model of balance of payments: The case ofVenezuela. Journal<br />
of Monetary <strong>Economic</strong>s 2, July, 311–332.<br />
Khan, MS and PJ Montiel (1989). Growth oriented adjustment programs. IMF Staff Papers<br />
36, June, 279–306.<br />
Khan, MS and PJ Montiel (1990), A marriage between fund and bank models. IMF Staff<br />
Papers 37, March, 187–191.<br />
Lazaridis, A (1980). Application of filtering methods in econometrics. International Journal<br />
of Systems Science 11(11), 1315–1325.<br />
Lucas, RE, Jr (1972). Expectation and the neutrality of money. Journal of <strong>Economic</strong> Theory<br />
4, April, 103–124.<br />
Meyer, T (1967). The lag in effect of monetary policy, some criticisms. Western <strong>Economic</strong><br />
Journal 5, September.<br />
Mayer, R (1972) Estimating coefficients that change over time. International <strong>Economic</strong><br />
Review 11, October.<br />
Muscatelli, VA and P Tirelli (2005). Analyzing the interaction of monetary and fiscal policy:<br />
does fiscal policy play a valuable role in stabilization. CESIFO <strong>Economic</strong> Studies 51(4),<br />
1–17.<br />
Nicolao, G (1992). On the time-varying Riccati difference equation of optimal filtering.<br />
SIAM Journal on Control and <strong>Optimization</strong> 30(6), November, 1251–1269.
Time Varying Responses 61<br />
Poole, W (1975). The relationship of monetary deceleration to business cycle peaks. Journal<br />
of Finance 30, June, 697–712.<br />
Radenkovic, M and A Michel (1992). Verification of the self-stabilization mechanism in<br />
robust stochastic adaptive control using Lyapunov function arguments. SIAM Journal<br />
on Control and <strong>Optimization</strong> 30(6), November, 1270–1294.<br />
Sargent, TJ and N Wallace (1973). Rational expectations and the theory of economic policy.<br />
Journal of Monetary <strong>Economic</strong>s 2, April, 169–183.<br />
Smets, V and R Wouters (2003). An estimated dynamic stochastic general equilibrium<br />
model of the Euro area. Journal of European <strong>Economic</strong> Association 1(5), September,<br />
1123–1175.<br />
Tobin, J (1970). Money and income: Post Hoc Ergo Propter Hoc. Quarterly Journal of<br />
<strong>Economic</strong>s 84, May, 301–317.<br />
Tsakalis, K and P Ioannou (1990). A new direct adaptive control scheme for time-varying<br />
plant. IEEE Transactions on Automatic Control 35(6) June, 356–369.<br />
Warburton, C (1971). Variability of the lag in the effect of monetary policy 1919–1965.<br />
Western <strong>Economic</strong> Journal 9, June, 1919–1965.<br />
Appendix A<br />
Probability Density Function and Recursive Process<br />
for Re-Estimation<br />
Under the assumptions stated in Section 3 and according to Bayes’ rule, the<br />
conditional probability density function of π i given x i+1 is Gaussian and is<br />
given by:<br />
∫<br />
p(π i+1 |x i+1 ) = p(π i |x i )p(π i+1 ,x i+1 |π i ,x i )dπ i<br />
where<br />
(<br />
p(π i+1 |x i ) = constant exp − 1 )<br />
2 ||π i+1 − πi ∗ ||2 Si<br />
−1<br />
πi ∗ = E(π i |x i )<br />
S i = cov(π i |x i )<br />
(S i assumed to be invertible)<br />
[<br />
P(π i+1 ,x i+1 |π i ,x i ) = constant exp − 1 ∥ π i+1 − π i ∥∥∥<br />
2<br />
2 ∥ x i+1 − H i+1 π i+1<br />
C =<br />
[ ] [<br />
Q1 0<br />
, C<br />
0 Q −1 Q<br />
−1<br />
=<br />
2<br />
1<br />
0<br />
0 Q −1<br />
2<br />
]<br />
.<br />
C −1 ]
62 Dipak R. Basu and Alexis Lazaridis<br />
Hence<br />
∫<br />
p(π i+1 |x i+1 ) = constant<br />
(<br />
exp − 1 )<br />
J<br />
2 ˜ i dπ i<br />
where<br />
J˜<br />
i = (π i − πi ∗ )′ (Si<br />
−1 + Q −1<br />
1 )(π i − πi ∗ )<br />
+ (π i+1 − πi ∗ )′ (Q −1<br />
1<br />
+ H i+1 ′ Q−1 2 H i+1)(π i+1 − πi ∗ )<br />
− 2(π i − πi ∗ )′ Q −1<br />
1 (π i − πi ∗ )<br />
+ (x i+1 − H i+1 πi ∗ )′ Q −1<br />
2 (x i+1 − H i+1 πi ∗ )<br />
− 2(π i+1 − πi ∗ )′ H i+1 ′ Q−1 2 (x i+1 − H i+1 πi ∗ ). (a)<br />
Now, define G −1<br />
i<br />
= (S −1<br />
i<br />
+ Q −1<br />
1<br />
) and consider the expression<br />
J˜<br />
i = [(π i − πi ∗ )′ − G i Q −1<br />
1 (π i+1 − πi ∗ )]′<br />
× G −1<br />
i<br />
[(π i − πi ∗ )′ − G i Q −1<br />
1 (π i+1 − πi ∗ )]. (b)<br />
Note that G i is symmetric, since it is the sum of two (symmetric) co-variance<br />
matrices. Expanding Eq. (b) and noting that G i G −1<br />
i<br />
= I we obtain<br />
J˜<br />
i = (π i − πi ∗ )′ G −1<br />
i<br />
(π i+1 − πi ∗ ) − 2(π i − πi ∗ )′ Q −1<br />
1 (π i+1 − πi ∗ )<br />
+ (π i+1 − πi ∗ )′ Q −1<br />
1 G iQ −1<br />
1 (π i+1 − πi ∗ ). (c)<br />
In view of Eqs. (b) and (c), Eq. (a) can be written as<br />
J˜<br />
i = [(π i − πi ∗ )′ − G i Q −1<br />
1 (π i+1 − πi ∗ )]′<br />
× G −1<br />
i<br />
[(π i − πi ∗ )′ − G i Q −1<br />
1 (π i+1 − πi ∗ )]<br />
+ (π i+1 − πi ∗ )′ (Q −1<br />
1<br />
+ H i+1 ′ Q−1 2 H i+1 − Q −1<br />
1 G iQ −1<br />
1 )(π i+1 − πi ∗ )<br />
+ (x i+1 − H i+1 πi ∗ )′ Q −1<br />
2 (x i+1 − H i+1 πi ∗ )<br />
− 2(π i+1 − πi ∗ )′ H i+1 ′ Q−1 2 (x i+1 − H i+1 πi ∗ ).<br />
∫ ( )<br />
Integration with respect to π i yields constant exp −<br />
1 ˜ 2<br />
J i dπi =<br />
constant exp ( − 1 ...<br />
3 J i)<br />
where<br />
...<br />
J i = ( π i+1 − πi ∗ )′ (Q −1<br />
1<br />
− Q −1<br />
1 G iQ −1<br />
1<br />
+ H i+1 ′ Q−1 2 H i+1) ( π i+1 − πi<br />
∗ )<br />
+ ( x i+1 − H i+1 πi ∗ )′ Q −1 (<br />
2 xi+1 − H i+1 πi<br />
∗ )<br />
− 2 ( π i+1 − πi ∗ )′ H i+1 ′ ( Q−1 2 xi+1 − H i+1 πi<br />
∗ )<br />
.
Time Varying Responses 63<br />
Hence<br />
(<br />
p(π i+1 |x i+1 ) = constant exp − 1 )<br />
...<br />
J i .<br />
2<br />
Since p(π i+1 |x i+1 ) is proportional to the likelihood function, by maximizing<br />
the conditional density function, we are also maximizing the likelihood<br />
in order to determine πi+1 ∗ ...<br />
. Note that minimization of J i, is equivalent to<br />
maximizing p(π i+1 |x i+1 ). To minimize ...<br />
J i, we expand Eq. (c) eliminating<br />
terms not containing π i+1 . Then, we differentiate with respect to π<br />
i+1 ′ and<br />
after equating to zero, we finally obtain (Lazaridis, 1980).<br />
πi+1 ∗ = π∗ i + (Q−1 1<br />
− Q −1<br />
1 G iQ −1<br />
1<br />
+ H i+1 ′ Q−1 2 H i+1) −1<br />
× H i+1 ′ Q−1 2 (x i+1 − H i+1 πi ∗ ). (d)<br />
Now, consider the composite matrix:<br />
Q −1<br />
1<br />
− Q −1<br />
1 G iQ1 −1 where G i = (Si<br />
−1 + Q −1<br />
1 )−1 .<br />
Considering the matrix identity of householder, we can write<br />
Q −1<br />
1<br />
− Q −1<br />
1 (S−1 i<br />
+ Q −1<br />
1 )−1 Q −1<br />
1<br />
= (Q 1 + S i ) −1 Pi+1 −1 .<br />
Hence Eq. (d) takes the form:<br />
πi+1 ∗ = π∗ i + K i+1(x i+1 − H i+1 πi ∗ )<br />
where K i+1 ,Si+1 −1 and P−1<br />
i+1<br />
are defined in Eqs. (21)–(23).<br />
It is recalled that H 0 is a null matrix, since no observations exist beyond<br />
period 1 in the estimation process. The same applies for the vector x 0 .<br />
Appendix B<br />
It is recalled that the model was estimated using FIML method. The FIML<br />
estimates are as follows (R 2 and adjusted R 2 refer to the corresponding 2<br />
SLS estimates. Numbers in brackets are the corresponding t-statistics).<br />
Estimated Model<br />
1. A t = 0.842Y t<br />
(3.48)<br />
+ 55191.5<br />
(4.87)<br />
+ 0.024A t−1 − 1.319IR t + 1.051IR t−t − 284EXR t<br />
(1.74) (1.34) (1.16) (1.29)<br />
R 2 = 0.99, R 2 = 0.92, DW = 1.76, ρ = 0.27
64 Dipak R. Basu and Alexis Lazaridis<br />
2. (Y t ) = A t + R t<br />
3. (BD) t = (G t + LR t + PF t ) − (TY t + GBS t + AF t + FB t )<br />
4. PF t = 1148.80 + 0.169 CFB t−1 + 144.374 WIR t<br />
(1.48) (2.39)<br />
(1.89)<br />
5. GBS t = 0.641G t<br />
(1.14)<br />
+ 0.677G t−1<br />
(1.41)<br />
+ 1.591IR t<br />
(1.64)<br />
− 0.33AF t<br />
(0.23)<br />
R 2 = 0.89, R 2 = 0.84, DW = 2.36, ρ = 0.23<br />
(1.51)<br />
6. MD = MS<br />
7. (MS) t = [(1 + CD t )/(CD t + RR t )](R t + NDA t )<br />
8. RR t = 0.008Y t<br />
(3.47)<br />
+ 0.065<br />
(2.72)<br />
9. CD t = 0.4391IR t<br />
(−1.49)<br />
10. MD t = 2.733RR t<br />
(−2.52)<br />
− 0.002Y t−1<br />
(−1.84)<br />
+ 1.2571R t<br />
(1.87)<br />
+ 2.7031R t−1<br />
(1.88)<br />
− 0.003T<br />
(2.87)<br />
R 2 = 0.86, R 2 = 0.79, DW = 2.88, ρ = 0.26<br />
(1.23)<br />
− 0.0057T<br />
(−6.269) + 0.193<br />
(11.95)<br />
+ 0.158CD t−1 + 0.0007Y t + 0.009Y t−1<br />
(1.16) (1.56) (1.73)<br />
R 2 = 0.94, R 2 = 0.92, DW = 1.22, ρ = 0.64<br />
(4.21)<br />
− 2.19IR t<br />
(−0.24)<br />
+ 1.713IR t−1 + 1.275Y t<br />
(2.17) (6.67)<br />
− 113335.0<br />
(−0.088)<br />
R 2 = 0.86, R 2 = 0.81, DW = 1.92, ρ = 0.26<br />
(1.31)<br />
11. IR t = 0.413MD t−1<br />
(1.93)<br />
+ 0.406Y t−1<br />
(1.56)<br />
12. P t = 0.0004A t<br />
(−5.71)<br />
− 0.814IR t−1 + 8.656CI t − 1.351CI t−1<br />
(1.403) (1.68)<br />
− 7.02<br />
(−1.84)<br />
R 2 = 0.98, R 2 = 0.95, DW = 2.48,ρ = 0.58<br />
(3.21)<br />
+ 0.0002A t−1 + 0.105IMC t<br />
(1.77)<br />
(1.48)<br />
+ 0.421P t−1<br />
(3.79)<br />
R 2 = 0.98, R 2 = 0.93, DW = 2.09,ρ = 0.48<br />
13. IMC t = 5.347 + 19.352WPM t + 9.017EXR t<br />
(1.34) (2.03)<br />
(0.97)<br />
R 2 = 0.98, R 2 = 0.97, DW = 2.07, ρ = 0.31<br />
(1.37)<br />
14. R t = X t − IM t + K t + PFT t + FB t − PF t + AF t<br />
+ 32.895<br />
(1.87)
15. IM t =−24.04 − 0.025IM t−1<br />
(−0.82)<br />
+ 1.123T<br />
(0.27)<br />
16. CFB t = ∑ t<br />
r=−20 FB r<br />
(−0.86)<br />
+ 0.104Y t<br />
(1.59)<br />
Time Varying Responses 65<br />
− 0.089Y t−1 − 0.654IMC t<br />
(9.21) (−1.26)<br />
R 2 = 0.96, R2 = 0.92, DW = 2.52, ρ =−0.62<br />
(−1.72)<br />
λ 2 = 563.09<br />
The estimated Chi-square satisfies the goodness of fit test for the FIML<br />
estimation.<br />
Notations<br />
A = Domestic absorption<br />
AF = Foreign receipts (grants etc.)<br />
BD = Government budget deficits<br />
CD = Credit to deposit ratio in the commercial banking sector<br />
CFB = Cumulative foreign borrowing, i.e., foreign debt over a period of<br />
20 years<br />
CT = Discount rate of the RBI<br />
FB = Foreign borrowing<br />
G = Government expenditure<br />
GBS = Government bond sales<br />
IM = Value of imports<br />
IMC = Import price index (1990 = 100)<br />
IR = Interest rate in the money-market<br />
K = Foreign capital inflows<br />
LR = Lending (minus repayments to the states)<br />
MD = Money demand<br />
MS = Money supply<br />
NDA = Net domestic asset creation by the RBI<br />
P = Consumers’ price index (1990 = 100)<br />
PFT = Private foreign transactions<br />
PF = Foreign payments<br />
R = Changes in foreign exchange reserve<br />
RR = Reserve to deposit ratio in the commercial banking sector<br />
TY = Government tax revenue<br />
T = Time trend
66 Dipak R. Basu and Alexis Lazaridis<br />
WPM = World price index of India’s imports (1990 = 100)<br />
WIR = World interest rate, average of European and US money market<br />
rate<br />
EXR = Exchange rate (Rs/US$)<br />
X = Value of exports<br />
Y = GNP at constant 1990 prices
Chapter 3<br />
Mathematical<br />
Modeling in Macroeconomics
This page intentionally left blank
Topic 1<br />
The Advantages of Fiscal Leadership in an<br />
Economy with Independent Monetary Policies<br />
Andrew Hughes Hallet<br />
James Mason University, USA<br />
1. Introduction<br />
In January 2006, Gordon Brown (as Britain’s finance minister) was widely<br />
criticized by the European Commission and other policy makers for running<br />
fiscal policies, which they considered to be too loose and irresponsible. The<br />
UK fiscal deficit had breached the 3% of GDP that is considered to be safe.<br />
To make its point, the European Commission initiated an “excessive deficit”<br />
procedure against the UK government, claiming that its fiscal policies constituted<br />
a danger for the good performance of the UK economy and its<br />
neighbors, if these deficits were not contained and reversed. Yet the United<br />
Kingdom, alone among its European partners, allows fiscal policy to lead in<br />
order to achieve certain social expenditure and medium-term output objectives<br />
and a degree of co-ordination with the monetary policies designed to<br />
reach a certain inflation target. And the economic performance has been<br />
no worse, and may have been better than elsewhere in the Eurozone. Was<br />
Gordon Brown’s strategy so mistaken after all?<br />
The British fiscal policy has changed radically since the days when it<br />
tried to micro-manage all of the aggregate demand with an accommodating<br />
monetary policy in the 1960s and 1970s; and again during the 1980s, it<br />
was designed to strengthen the economy’s supply-side responses, while the<br />
monetary policies were intended to secure lower inflation.<br />
The 1990s saw a return to more activist fiscal policies — but policies<br />
designed strictly in combination with an equally active monetary policy<br />
based on inflation targeting and an independent Bank of England. They are<br />
set, in the main, to gain a series of medium- to long-term objectives — low<br />
debt, the provision of public services and investment, social equality, and<br />
economic efficiency. The income stabilizing aspects of the fiscal policy<br />
69
70 Andrew Hughes Hallet<br />
have, therefore, been left to act passively through the automatic stabilizers,<br />
which are part of any fiscal system, while the discretionary part (the bulk of<br />
the policy measures) is set to achieve these long-term objectives. Monetary<br />
policy, meanwhile, is left to take care of any short-run stabilization around<br />
the cycle; that is, beyond what, predictably, would be done by the automatic<br />
stabilizers. 1<br />
To draw a sharp distinction between actively managed long-run policies,<br />
and non-discretionary short-run stabilization efforts restricted to the<br />
automatic stabilizers, is of course the strategic policy prescription of Taylor<br />
(2000). Marrying this with an activist, monetary policy directed at cyclical<br />
stabilization, but based on an independent Bank of England and a monetary<br />
policy committee with the instrument (but not target) independence,<br />
appears to have been the innovation in the UK policies. It implies a leadership<br />
role for the fiscal policy, which allows both fiscal and monetary<br />
policies to be better co-ordinated — but without either losing their ability<br />
to act independently. 2<br />
Thus, Britain appears to have adopted a Stackelberg solution, which lies<br />
somewhere between the discretionary (but Pareto superior) co-operative<br />
solution, and the inferior but independent (non-co-operative) solution, as<br />
shown in Fig. 1. 3 Nonetheless, by forcing the focus onto long-run objectives,<br />
to the exclusion of the short term, this set-up has imposed a degree of<br />
pre-commitment (and potential for electoral punishment) on fiscal policy<br />
because governments naturally wish to lead. But, the regime remains nonco-operative<br />
so that there is no incentive to renege on earlier plans in the<br />
1 The Treasury estimates that the automatic stabilizers will, in normal circumstances, stabilize<br />
some 30% of the cycle; the remaining 70% being left to monetary policy (HM Treasury,<br />
2003). The option to undertake discretionary stabilizing interventions is retained “for<br />
exceptional circumstances” however. Nevertheless, the need for any such additional interventions<br />
is unlikely: first, because of the effectiveness of the forward looking, symmetric<br />
and an activist inflation targeting mechanism adopted by the Bank of England; and, second,<br />
because the long-term expenditure (and tax) plans are deliberately constructed in nominal<br />
terms so that they add to the stabilizing power of the automatic stabilizers in more serious<br />
booms or slumps.<br />
2 For details on how this leadership vs. stabilization assignment is intended to work, see HM<br />
Treasury (2003) and Section 3 below. Australia and Sweden operate rather similar regimes.<br />
3 Figure 1 is the standard representation of the outcomes of the different game theory equilibria<br />
when the reaction functions form an acute angle. The latter is assured for our context<br />
when both sets of policy makers have some interest in inflation control and output stabilization,<br />
and the policy multipliers have the conventional signs (Hughes Hallett and Viegi,<br />
2002). Stackelberg solutions, therefore, imply a degree of co-ordination in the outcomes,<br />
relative to the non-co-operative (unco-ordinated) Nash solution.
The Advantages of Fiscal Leadership in an Economy 71<br />
Figure 1.<br />
Monetary-Fiscal Interactions and the Central Bank.<br />
absence of changes in information. Thus, the policies and their intended<br />
outcomes will be sustained by the government of the day. 4<br />
2. The Fiscal Leadership Hypothesis<br />
The hypothesis is that the United Kingdom’s improved performance is due<br />
to the fact that the fiscal policy leads an independent monetary policy. This<br />
leadership derives from the fact that fiscal policies typically have long-run<br />
targets (sustainability and low debt), and is not easily reversible (public services<br />
and social equality), and does not stabilize well if efficiency is to be<br />
maintained. Nevertheless, there are also automatic stabilizers in any fiscal<br />
policy framework, implying that monetary policy must condition itself on<br />
the fiscal stance at each point. This automatically puts the latter in a follower’s<br />
role. This is helpful because it allows the economy to secure the benefits<br />
of an independent monetary policy but also enjoys a certain measure<br />
of co-ordination between the two sets of policy makers — discretionary/<br />
automatic fiscal policies on one side, and active monetary policies on the<br />
4 Stackelberg games, with fiscal policy leading, produce fiscal commitment: i.e., sub-game<br />
perfection with either strong or weak time consistency (Basar, 1989). In this sense, we have<br />
“rules rather than discretion”. Commitment to the stabilization policies is then assured by<br />
the independence of the monetary authority.
72 Andrew Hughes Hallet<br />
other. The extra co-ordination arises in this case because the constraints<br />
on, and responses to an agreed leadership role reduce the externalties of<br />
self-interested behavior, which independent agents would otherwise impose<br />
on one another. This allows a Pareto improvement over the conventional<br />
non-co-operative (full independence) solution, without reducing the central<br />
bank’s ability to act independently. It is important to realize that the<br />
co-ordination here is implicit and therefore rule-based, not discretionary.<br />
To show that the UK fiscal policy does lead monetary policy in this<br />
sense, and that this is the reason for the Pareto improved results in Table 1,<br />
I produce evidence in three parts:<br />
• Institutional evidence: taken from the UK Treasury’s own account of<br />
how its decision-making works, what goals it needs to achieve, and how<br />
the monetary stance may affect it or vice versa;<br />
Table 1.<br />
Generalized Taylor rules in the UK and EU-12.<br />
S.No. Const r t−1<br />
π j,k<br />
e<br />
j, k gap pd debt<br />
Dependent variable: central bank lending rate, r t :<br />
For the UK, monthly data from June 1997–January 2004<br />
(1) −1.72 0.711 1.394 +6, +18 0.540<br />
(2.16) (5.88) (2.66)<br />
(2.06)<br />
(2) −2.57<br />
(2.59)<br />
0.598<br />
(4.38)<br />
R 2 = 0.91, F 3,21 = 82.47, N = 29<br />
1.289<br />
(0.66)<br />
+9, +21 1.10<br />
(1.53)<br />
— —<br />
−0.67<br />
(0.83)<br />
0.43<br />
(0.50)<br />
R 2 = 0.90, F 5,19 = 44.1, N = 25<br />
For the Eurozone (EU-12), monthly data from January 1999–January 2004<br />
(1) −0.996<br />
(0.76)<br />
(2) −13.67<br />
(0.59)<br />
0.274<br />
(1.04)<br />
0.274<br />
(1.04)<br />
1.714<br />
(3.89)<br />
+9, +21 0.610<br />
(1.77)<br />
R 2 = 0.82, F 3,11 = 22.2, N = 15<br />
1.110<br />
(1.19)<br />
−6, +6 0.341<br />
(1.22)<br />
— —<br />
0.463<br />
(2.89)<br />
R 2 = 0.87, F 5,13 = 24.7, N = 19<br />
0.191<br />
(0.63)<br />
Notation: j, k give the bank’s inflation forecast interval; πj,k e represents the average inflation<br />
rate expected over the interval t + j to t + k: Eπ t+j,t+k ;gap= GDP–trend GDP; pd =<br />
primary deficit/surplus as a percentage of GDP (a surplus > 0) and debt = debt/GDP ratio.<br />
Estimation: instrumented 2SLS; t-ratios in parentheses; j, k determined by search; and the<br />
output gap is obtained from a conventional HP filter to determine trend output.
The Advantages of Fiscal Leadership in an Economy 73<br />
• Empirical evidence: the extent to which the monetary policy is affected<br />
by the fiscal policy, but fiscal policy with its longer objectives does not<br />
depend on monetary policies and<br />
• Theoretical evidence: shows how, with different goals for governments<br />
and central banks, Pareto improving results can be expected from fiscal<br />
leadership. The United Kingdom has, therefore, had the incentive, and<br />
the capacity, to operate in this way.<br />
3. Institutional Evidence: The Treasury’s Mandate<br />
3.1. History<br />
The British fiscal policy has changed a great deal over the past 30 years.<br />
Fiscal policy was the principal instrument of the economic policy in the<br />
1960s and 1970s, and was focused almost exclusively on demand management.<br />
Monetary policy and the provision of public services (subject to a<br />
lower bound) were essentially accommodating factors, since the main constraint<br />
was perceived to be the financing of persistent current-account trade<br />
deficits. The result of this was a short-term view, in which the conflicts<br />
between the desire for growth (employment) and the recurring evidence<br />
of overheating (trade deficits) led to an unavoidable sequence of “stop-go”<br />
policies. There were many (Dow, 1964), who argued that the real problem<br />
was a lack of long-run goals for the fiscal policy; and that the lags inherent<br />
in recognizing the need for a policy change and in implementing it through<br />
Parliament till it takes hold in the markets, had produced a system that was<br />
actually destabilizing rather than stabilizing. But, there would have been<br />
conflicts anyway because there were two or more competing goals, but<br />
effectively only one instrument to reach them.<br />
Such a system could not provide the long-run goal of public services,<br />
public investment, and social equity. It certainly proved too difficult to turn<br />
the fiscal policy on and off as fast and frequently as required, and precommit<br />
to certain expenditure and taxation plans at the same time. The<br />
point here is that, if you cannot pre-commit fiscal policy to certain goals,<br />
then you will not be able to pre-commit monetary policy either. This means<br />
that the “stable growth with low inflation” objective will be lost.<br />
In the 1980s, the strategy changed when a new regime took office, committed<br />
to reducing the size and role of the government in the economy. The<br />
demand management role of the fiscal policy was phased out. This role<br />
passed over to inflation control and monetary management — with limited
74 Andrew Hughes Hallet<br />
success in the earlier phases of monetary and exchange rate targeting, but<br />
with rather more success when it was formalized as an inflation-targeting<br />
regime under an independent Bank of England and monetary policy committee.<br />
In this period, therefore, the role of fiscal policy was to provide<br />
the “conditioning information” within which the monetary management<br />
had to work. It was split into two distinct parts. Short-term stabilization<br />
of output and employment was left to the automatic stabilizers inherent in<br />
the prevailing tax and expenditures rules. Discretionary adjustments would<br />
only be used in exceptional circumstances, if at all. 5 The rest of the fiscal<br />
policy, being the larger part, could then be directed at long-term objectives:<br />
in this instance, changes to free up the supply side (and enhance competitiveness)<br />
on the premise that greater microeconomic flexibility would<br />
both reduce the need for stabilizing interventions (relative wage and price<br />
adjustments would do the job better) and provide the conditions for stronger<br />
employment and growth (HMT, 2003). Given this, the long-run goals of<br />
better public services, efficiency, and social equity could then be attained.<br />
In addition, as the market flexibility changes were made, this part of fiscal<br />
policy could be increasingly focused on providing the long-run goals<br />
directly. 6<br />
3.2. Current Practice<br />
According to the treasury’s own assessment (HMT, 2003), the UK fiscal policy<br />
now leads monetary policy in two senses. First, fiscal policy is decided<br />
in advance of monetary or other policies (pp. 5, 63, 64, 74), 7 and with a<br />
long-time horizon (pp. 5, 42, 48, 61). Second, monetary policy is charged<br />
with controlling inflation and stabilizing output around the cycle — rather<br />
than steering the economy as such (pp. 61–63). Fiscal policy, therefore, sets<br />
the conditions within which monetary policy has to respond and achieve its<br />
own objectives (pp. 9, 15, 67–68). The short-term fiscal interventions, now<br />
less than half the total and declining (p. 48, Box 5.3, Section 6), are restricted<br />
to the automatic stabilizers — with effects that are known and predictable.<br />
5 This follows the recommendations in Taylor (2000). However, because of the inevitable<br />
trade-off between cyclical stability and budget stability, it is likely to be successful only in<br />
markets with sufficiently flexible wages and prices (see Fatas et al., 2003).<br />
6 This summary is taken from the UK Treasury’s own view (HMT, 2003), Sections 2, 4, and<br />
pp. 34–38.<br />
7 In this section, page or section numbers given without further reference are all cited from<br />
HMT (2003).
The Advantages of Fiscal Leadership in an Economy 75<br />
The short-run discretionary components are negligible (p. 59, Table 5.5),<br />
and the long-run objectives of the policy will always take precedence in<br />
cases of conflict (pp. 61, 63–68). Fiscal policy would, therefore, not be<br />
used for fine tuning (pp. 11, 63), or for stabilization (pp. 1, 14). This burden<br />
will be carried by interest rates, given the fiscal stance and its forward plans<br />
(pp. 1, 7, 11, 37).<br />
Third, given the evidence that consumers and firms often do not look<br />
forward efficiently and may be credit-constrained, and also that the impacts<br />
of the fiscal policy are uncertain and have variable lags (pp. 19, 26, 48;<br />
Taylor, 2000), it makes sense for the fiscal policy to be used consistently<br />
and sparingly and in a way that is clearly identified with the long-term<br />
objectives. The United Kingdom has determined these objectives to be (pp.<br />
11–13, 39–41, 61–63, 81–82):<br />
(a) The achievement of sustainable public finances in the long term; low<br />
debt (40% of GDP) and symmetric stabilization over the cycle.<br />
(b) A sustainable and improving delivery of public services (health and education);<br />
improving competitiveness/supply-side efficiency in the long<br />
term; and the preservation of social (and inter-generational) equity and<br />
alleviation of poverty.<br />
(c) Recognition that the achievement of these objectives is often contractual<br />
and cannot easily be reversed once committed. The long lead times<br />
needed to build up these programs mean that the necessary commitments<br />
must be made well in advance. Frequent changes would conflict<br />
with the government’s medium-term sustainability objectives, and cannot<br />
be fulfilled. Similarly, changes in direct taxes will affect both equity<br />
and efficiency, and should seldom be made.<br />
(d) The formulation of clear numerical objectives is consistent with these<br />
goals; a transparent set of institutional rules to achieve them; and a commitment<br />
to a clear separation of these goals from economic stabilization<br />
around the cycle. 8<br />
(e) To ensure that fiscal policy can operate along these lines, public expenditures<br />
are planned with fixed three-year departmental expenditure limits<br />
which, when combined with decision and implementation lags of up<br />
to two years, means that the bulk of the fiscal policy has to be planned<br />
8 The credibility of fiscal policy, and by extension an independent monetary policy, is seen as<br />
depending on these steps, and on symmetric stabilization. See also Dixit (2001), and Dixit<br />
and Lambertini (2003).
76 Andrew Hughes Hallet<br />
with a horizon of up to five years — vs a maximum of two years in<br />
the inflation targeting rules operated by the Bank of England. Moreover,<br />
these spending limits are constraints defined in nominal terms, so<br />
that they will be met. In addition, the Treasury operates a tax smoothing<br />
approach — having rejected “formulaic rules” that might have<br />
adjusted taxes or expenditures, or the various tax regulators, or credit<br />
taxes that could have stabilized the economy in the short term. The<br />
bulk of fiscal policy will remain focused on the medium to long term<br />
therefore.<br />
To summarize, the institutional structure itself has introduced fiscal<br />
leadership. And, since the discretionary elements are smaller than elsewhere<br />
— smaller than the automatic stabilizers (Tables 1–5) and declining<br />
(Box 5.3) — something else (monetary policy) must have taken up the burden<br />
of short-run stabilization. This arrangement is, therefore, best modeled<br />
as a Stackelberg game, with institutional parameters for goals, independence,<br />
priorities etc. I shall argue that this has had the very desirable result<br />
of increasing the degree of co-ordination between policy makers without<br />
compromising their formal or institutional independence — or indeed, their<br />
credibility for delivering stability and low inflation. Constrained independence,<br />
in other words, designed to reduce the externalities, which each party<br />
might, otherwise, impose on the other.<br />
4. Empirical Evidence<br />
The next step is to establish whether the UK authorities have followed<br />
the leadership model described above. The fact that they say that they<br />
follow such a strategy does not prove that they do so. The difficulty<br />
here is that, although the asymmetry in ex-ante (anticipated) responses<br />
between the follower and leader is clear enough in the theoretical model —<br />
the follower expects no further changes from the leader after the follower<br />
chooses his reaction function, whereas the leader takes this reaction<br />
function into account — such an asymmetry and zero restriction<br />
will not appear in the ex-post (observed) responses that emerge in the<br />
final solution (Basar and Olsder, 1999; Brandsma and Hughes Hallett,<br />
1984; Hughes Hallett, 1984). Since I have no data on anticipations, this<br />
makes it difficult to test for leadership directly. But, we can use indirect<br />
tests based on the degree of competition, or complimentarity, between the<br />
instruments.
4.1. Monetary Responses<br />
The Advantages of Fiscal Leadership in an Economy 77<br />
For monetary policy, it is widely argued that the authorities’ decisions can<br />
best be modeled by a Taylor rule 9 :<br />
r t = ρr t−1 + αE t π t+k + βgap t+h α, β, ρ ≥ 0 (1)<br />
where k, h represent the authorities’forecast horizon 10 and may be positive<br />
or negative. Normally, α>1 will be required to avoid indeterminancy:<br />
that is, arbitrary variations in output or inflation, as a result of unanchored<br />
expectations in the private sector. The relative size of α and β then reveals<br />
the strength of the authorities’ attempts to control inflation vs. income stabilization;<br />
and ρ their preference for gradualism. I set h = 0 in Eq. (1),<br />
since monetary policy appears not to depend on the expected output gaps<br />
(Dieppe et al., 2004).<br />
In order to obtain an idea of the influence of fiscal policies on monetary<br />
policy, I include some Taylor rule estimates — with and without fiscal<br />
variables — in Table 1. They show such rules for the United Kingdom and<br />
the Eurozone since 1997 and 1999, respectively, the dates when new policy<br />
regimes were introduced. The Eurozone has been included to emphasize<br />
the potential contrast between fiscal leadership in the United Kingdom, and<br />
the lack of it in Europe.<br />
4.2. Different Types of Leadership<br />
Conventional wisdom would suggest that Europe has either monetary leadership<br />
or independent policies, and hence policies which are either jointly<br />
dependent in the usual way, or which are complementary and mutually supporting.<br />
The latter implies that the monetary policy tends to expand/contract<br />
whenever fiscal policy needs to expand or contract — but not necessarily<br />
vice versa, when money is expanding or contracting and is sufficient to<br />
9 Taylor (1993b). One can argue that the policy should be based on fully optimal rules<br />
(Svensson, 2003), of which Eq. (1) will be a special case. But, it is hard to argue that policy<br />
makers actually do optimize when the additional gains from doing so may be small and when<br />
the uncertainties in their information, policy transmissions, or the economy’s responses may<br />
be quite large. In practice, therefore, the Taylor rule approach is often found to fit central<br />
bank behavior better.<br />
10 In principle, k and h may be positive or negative: positive if the policy rule is based on<br />
future-expected inflation, to head off an anticipated problem, as in the Bank of England. But<br />
negative if interest rates are to follow a feedback rule to correct past mistakes or failures.
78 Andrew Hughes Hallet<br />
control inflation on its own. This is a weak form of monetary leadership in<br />
which fiscal policies are an additional instrument for use in cases of particular<br />
difficulty, rather than the policies in a Nash game with conflicting<br />
aims that need to be reconciled.<br />
More generally, leadership implies complimentarity among policy<br />
instruments in the leader’s reaction function, but conflicts among them in<br />
the follower’s responses. A weak form of leadership also allows for independence<br />
among instruments in the leader’s policy rules. Thus, monetary<br />
leadership would imply some complimentarity (or independence) in the<br />
Taylor rule, but conflicts in the fiscal responses.And fiscal leadership would<br />
mean complimentarity or independence in the fiscal rule, but conflicts in<br />
the monetary responses. Evidently, from Section 3, we might expect Stackelberg<br />
leadership (with fiscal policy leading) in the UK, but the opposite in<br />
the Eurozone.<br />
4.3. Observed Behavior<br />
The upper equations in each panel of Table 1 yield the standard results for<br />
the monetary behavior in both the United Kingdom and the Eurozone. Both<br />
monetary authorities have targeted expected inflation more than the output<br />
gap since the late 1990s — and with horizons of 18–21 months ahead. The<br />
European Central Bank (ECB) has been more aggressive in this respect.<br />
But, contrary to conventional wisdom, it was also more sensitive to the<br />
output gap and had a longer horizon and less policy inertia.<br />
However, if we allow monetary policies to react to the changes in fiscal<br />
stance, we get different results (the lower equations). Here, we see that the<br />
UK monetary decisions may take fiscal policy into account, but the effect is<br />
not significant or well defined. However, this model of monetary behavior<br />
does imply more activist policies, a longer forecast horizon (up to two years<br />
as the Bank of England claims) and greater attention to the output gap —<br />
the symmetry in the UK’s policy rule. And to the extent that fiscal policy<br />
does have an influence, it would be as a substitute (or competitor) for monetary<br />
policy — fiscal deficits lead to higher interest rates. This is potentially<br />
consistent with fiscal leadership, since this form of leadership can allow<br />
independence or complimentarity between instruments in the leader’s rule.<br />
We need to check the fiscal reaction functions directly. The lack of significance<br />
for the debt ratio is easily understood, however. Since this is a<br />
declared long-run objective of the fiscal policy, it would not be necessary<br />
for the monetary policy to take it into account. So far, the evidence could
The Advantages of Fiscal Leadership in an Economy 79<br />
imply a Stackelberg follower role or independence for monetary policy,<br />
easing the competition (externalities) among policy instruments.<br />
The ECB results look quite different. Once fiscal effects are included,<br />
the concentration on inflation control is much reduced and the forecast<br />
horizon shrinks to six months. Moreover, a feedback element, to correct<br />
the past mistakes, comes in. At the same time, output stabilization becomes<br />
less important, which implies that the symmetric targeting goes out. Instead,<br />
monetary policy now appears to react to fiscal policy, but with the “wrong”<br />
sign: the larger the primary deficit, the looser will be the monetary policy.<br />
In this case, therefore, the policies are acting as complements — circumstances,<br />
which call for a primary deficit will also call for a relaxation of the<br />
monetary policy. Thus, we have an evidence of monetary leadership, where<br />
fiscal policy is used as an additional policy instrument.<br />
4.4. Fiscal Reaction Functions in the UK and Eurozone<br />
Many analysts have hypothesized that fiscal policy responses can best be<br />
modeled by means of a “fiscal Taylor rule”:<br />
d t = a t + γgap t + sd γ > 0 (2)<br />
where sd = structural deficit ratio, d t is the actual deficit ratio (d >0<br />
denotes a surplus), 11 and a t represents all the other factors such as the<br />
influence of monetary policy, existing or anticipated inflation, the debt burden,<br />
or discretionary fiscal interventions. The coefficient γ then gives a<br />
measure of an economy’s automatic stabilizers. The European Commission<br />
(2002), for example, estimates γ ≈ 1/2 for Europe — a little more in<br />
countries with an extensive social security system, a little less elsewhere.<br />
And a similar relationship is thought to underlie UK’s fiscal policy (see<br />
HMT, 2003: Boxes 5.2 and 6.2).<br />
The expected signs of any remaining factors are not so clear. The debt<br />
burden should increase current deficits, unless there is a systematic debt<br />
reduction program underway. Inflation should have a positive impact on the<br />
deficit ratio if fiscal policy is used for stabilization purposes, but it had no<br />
effect otherwise. Finally, the output gap should also have a positive impact<br />
on the deficit ratio, if the latter is being used for stabilization purposes —<br />
in which case interest rates should be negatively correlated with the size<br />
11 See Taylor (2000), Gali and Perotti (2003), Canzoneri (2004), and Turrini and in ‘t Veld<br />
(2004) for similar formulations. The European Commission (2002) uses a similar rule, but<br />
defines d>0 to be a deficit. They therefore expect γ to be negative.
80 Andrew Hughes Hallet<br />
Table 2.<br />
Fiscal policy reaction functions.<br />
For the United Kingdom, sample period 1997q3–2004q2<br />
d t =−0.444 + 0.845debt t + 0.0685r t + 1.076gap t<br />
(0.37) (7.64) (0.30) (1.72)<br />
For the Eurozone, sample period 1999q1–2004q2<br />
d t = 3.36 + 1.477d t−1 − 0.740π<br />
(2.96)<br />
+9,21<br />
e − 0.207r t − 0.337gap t<br />
(9.66) (2.21) (1.16) (2.18)<br />
R 2 = 0.78,<br />
F 3,24 = 33.23<br />
R 2 = 0.95,<br />
F 4,11 = 54.55<br />
Key: d t = gross deficit/GDP (%), where d
The Advantages of Fiscal Leadership in an Economy 81<br />
Consequently, the United Kingdom appears to use fiscal policy for stabilization<br />
in a minor way as claimed, while the Euro economies seem not<br />
to use fiscal policy this way at all. Confirmation of this comes from the<br />
responses to interest rate changes, which are positive but very small and<br />
statistically insignificant in the United Kingdom, but negative and near significant<br />
in Europe. This implies that the British fiscal policies are chosen<br />
independently of the monetary policy, as suggested by our weak leadership<br />
model. But, if there is any association at all, then both the fiscal and<br />
monetary policies would be compliments and weakly co-ordinated.<br />
The UK results are therefore inconclusive. They are consistent with<br />
independent policies, or fiscal leadership — the latter, despite the insignificance<br />
of the direct fiscal-monetary linkage, because of the significant output<br />
gap term in the fiscal equation which, given the same effect is not found in<br />
the monetary policy reactions, suggests fiscal leadership may in fact have<br />
been operating. In any event, there is no suggestion of monetary leadership;<br />
if anything happens, the results imply independence or fiscal leadership.<br />
In the Eurozone, the results are quite different. Here, the significant<br />
result is the conflict among instruments in monetary policy (and essentially<br />
the same conflict in the fiscal policy reactions). Hence, the policies are<br />
competitive, which suggests that they form a Nash equilibrium (or possibly<br />
monetary leadership, since the coefficient on fiscal policy in the Taylor<br />
rule is small and there is no output gap smoothing). This suggests weak<br />
monetary leadership, or a simple non-co-operative game.<br />
5. Theoretical Evidence: A Model of Fiscal Leadership<br />
5.1. The <strong>Economic</strong> Model and Policy Constraints<br />
The key question now is: would governments want to pursue fiscal precommitment?<br />
Do they have an incentive to do so? And would there be a<br />
clear improvement in terms of an economic performance if they did? More<br />
important, would the fiscal leadership model be more advantageous, if fiscal<br />
policy was limited by a deficit rule in the form of “hard” targets (as in the<br />
original stability pact) or “soft” targets?<br />
To answer these questions, we extend a model used in Hughes Hallett<br />
and Weymark (2002; 2004a;b; 2005) to examine the problem of monetary<br />
policy design when there are interactions with fiscal policy. For exposition<br />
purposes, we suppress the spillovers among countries and focus on the<br />
following three equations to represent the economic structure of any one
82 Andrew Hughes Hallet<br />
country 13 :<br />
π t = π e t + αy t + u t (3)<br />
y t = β(m t − π t ) + γg t + ε t (4)<br />
g t = m t + s(by t − τ t ) (5)<br />
where π t is inflation in period t, y t is the growth in output (relative to trend)<br />
in period t, and πt e represents the rate of inflation that the rational agents<br />
expect to prevail in period t, conditional on the information available at the<br />
time expectations are formed. Next, m t ,g t , and τ t represent the growth in<br />
the money supply, government expenditures, and tax revenues in period t,<br />
and u t and ε t are random disturbances, which are distributed independently<br />
with zero mean and constant variance. All variables are deviations from<br />
their steady-state (or equilibrium) growth paths, and we treat trend budget<br />
variables as pre-committed and balanced. Deviations from the trend budget<br />
are, therefore, the only discretionary fiscal policy choices available. The<br />
coefficients α, β, γ, s, and b are all positive by assumption. The assumption<br />
that γ is positive is sometimes controversial. 14 However, the short-run<br />
impact multipliers derived from Taylor’s (1993a) multicountry estimation<br />
provide an empirical support for this assumption (as does the HMT, 2003).<br />
According to Eq. (3), inflation is increasing, as the rate of inflation predicted<br />
by private agents and in output growth. Equation (4) indicates that<br />
both monetary and fiscal policies have an impact on the output gap. The<br />
micro-foundations of the aggregate supply Eq. (3), originally derived by<br />
Lucas (1972; 1973), are well known. McCallum (1989) shows that aggregate<br />
demand equation like Eq. (4) can be derived from a standard, multiperiod<br />
utility-maximization problem.<br />
Equation (5) describes the government’s budget constraint. In earlier<br />
chapters, we allowed discretionary tax revenues to be used for<br />
13 Technically, we assume a blockwise orthogonalization of the traditional multicountry<br />
model to produce independent semi-reduced forms for each country. The disturbance terms<br />
may, therefore, contain foreign variables, but they will have zero means so long as these<br />
countries remain on their long-run (equilibrium) growth paths on average (all variables being<br />
defined as deviations from their equilibrium growth paths).<br />
14 Barro (1981) argues that government purchases have a contractionary impact on output.<br />
Our model, by contrast, treats fiscal policy as important because: (i) fiscal policy is widely<br />
used to achieve re-distributive and public service objectives; (ii) governments cannot precommit<br />
monetary policy with any credibility if fiscal policy is not pre-committed (Dixit<br />
and Lambertini, 2003), and (iii) Central Banks, and the ECB, in particular, worry intensely<br />
about the impact of fiscal policy on inflation and financial stability (Dixit, 2001).
The Advantages of Fiscal Leadership in an Economy 83<br />
re-distributive purposes only, but permitted discretionary expenditures for<br />
enhancing output.We further assumed that there were two types of agents —<br />
rich and poor — and that only the rich use their savings to buy government<br />
bonds. Based on this view, b would be the proportion of pre-tax income<br />
(output) going to the rich and s, the proportion of after-tax income that the<br />
rich allocate to saving. Tax revenues, τ t , can then be used by the government<br />
to re-distribute income from the rich to poor, either directly or via public<br />
services. This structure, therefore, has output-enhancing expenditures g t ,<br />
and discretionary transfers τ t . Both are financed by aggregate tax revenues;<br />
that is, from discretionary and trend revenues. Expenditures above these<br />
revenues must be financed by the sale of bonds.<br />
5.2. An Alternative Interpretation<br />
We could, however, take a completely different interpretation of Eq. (5).<br />
We could take the term s(by t − τ t ) to be the proportion of the budget-deficit<br />
ratio, as it currently exists, adjusted for the effect of growth on this ratio<br />
and any discretionary taxes raised in this period, which the government<br />
now proposes to spend in period t. The deficit ratio itself is of course<br />
d = (G − T)/Y = (e − r), where G and T are the absolute levels of<br />
government spending and tax revenues, and e and r are their counterparts as<br />
a proportion of Y. If the economy grows, then d = (G−T)/Y −ẏ(e−r)<br />
where ẏ denotes the growth rate in national income. If we wish to see what<br />
would happen to this deficit ratio if fiscal policies were not changed, it<br />
means new spending can only be allowed to take place out of new revenues<br />
generated by growth: G = T = rY. Inserting this, we have d =<br />
−(e − r)ẏ =−dẏ. Since y t is the deviation of Y from its steady-state<br />
path, by t in Eq. (5) will equal d, the change in the existing deficit ratio<br />
under existing expenditure plans and tax codes, if b =−(e − r)/Y 1 . The<br />
government will spend some proportion of that, s, less new discretionary<br />
taxes in the current period. Hence, s would typically equal 1, although it<br />
might be less, if some parts were saved in social security funds or other<br />
assets. This defines what would happen to the deficit ratio if there were no<br />
change in existing fiscal policies; but the term in τ t shows that governments<br />
may also raise additional taxes in order to reduce the current deficit ratio<br />
to some desired target value, θ, say. This target is likely to be θ = 0, to<br />
balance the budget over the cycle, as required by the stability pact. We give<br />
governments such a target, in the form of either a soft or a hard rule, in the<br />
objective functions that are discussed in Section 5.
84 Andrew Hughes Hallet<br />
Given this new interpretation, we can now solve for πt e,π t, and y t from<br />
Eqs. (3) and (4). This yields the following reduced forms:<br />
[<br />
π t (g t ,m t ) = (1 + αβ) −1 αβm t + αγg t + m e t + γ ]<br />
β ge t + αε t + u t (6)<br />
y t (g t ,m t ) = (1 + αβ) −1 [βm t + γg t − βm e t − γge t + ε t − βu t ]. (7)<br />
Solving for τ t using Eqs. (5) and (7), then yields:<br />
τ t (g t ,m t ) = [s(1 + αβ)] −1 [(1 + αβ + sbβ)m t − (1 + αβ − sbγ)g t<br />
− sbβm e t − sbγge t + sb(ε t − βu t )]. (8)<br />
5.3. Government and Central Bank Objectives<br />
In our formulation, we allow for the possibility that the government and<br />
an independent central bank may differ in their objectives. In particular,<br />
we assume that the government cares about inflation stabilization, output<br />
growth, and the provision of public services (and hence the size of the public<br />
sector deficit or debt); whereas the central bank, if left to itself, would be<br />
concerned only with the first two objectives, 15 and possibly only the first<br />
one. We also assume that the government has been elected by majority<br />
vote, so that the government’s loss function reflects society’s preferences<br />
to a significant extent.<br />
Formally, the government’s loss function is given by:<br />
L g t = 1 2 (π t −ˆπ) 2 − λ g 1 y t + λg 2<br />
2 [(b − θ)y t − τ t ] 2 (9)<br />
where ˆπ is the government’s inflation target, λ g 1<br />
is the relative weight or<br />
importance that the government assigns to output growth, 16 and λ g 2<br />
is the<br />
relative weight, which it assigns to the debt or deficit rule. The parameter θ<br />
represents the target value for the debt or deficit to the GDP ratio, which the<br />
government would like to reach: hence (b − θ)y t becomes the target for its<br />
15 Since the central bank has no instruments to control the debt itself, it can only react to<br />
poor fiscal discipline indirectly: e.g., to the extent that its inflation objective is compromised,<br />
where it does have an instrument.<br />
16 Barro and Gordon (1983) also adopt a linear output target. In the delegation literature, the<br />
output component in the government’s loss function is usually represented as quadratic to<br />
reflect an output stability objective. In our model, the quadratic term in debt/deficits allows<br />
monetary and fiscal policy to play a stabilization role as well as pick a position on the<br />
economy’s output-inflation trade-off.
The Advantages of Fiscal Leadership in an Economy 85<br />
discretionary revenues, τ t . All the other variables are as previously defined.<br />
We may regard large values of λ g 2 , say λg 2 > max[1,λg 1<br />
], as defining a<br />
“hard” debt or deficit rule; and smaller values, λ g 2 < min[1,λg 1<br />
], as a “soft”<br />
debt or deficit rule.<br />
The objectives of the central bank, however, may be quite different from<br />
those of the government. We model them as follows:<br />
L cb<br />
t<br />
= 1 2 (π −ˆπ)2 − (1 − δ)λ cb y t − δλ g 1 y t<br />
+ δλg 2<br />
2 [(b − θ)y t − τ t ] 2 (10)<br />
where 0 ≤ δ ≤ 1, and λ cb is the weight, which the central bank assigns<br />
to the output growth. The parameter δ measures the degree to which the<br />
central bank is forced to take the government’s objectives into account.<br />
The closer δ is to 0, the greater is the independence of the central bank in<br />
making its choices. And the lower is λ cb is, the greater is the degree of its<br />
conservatism in making these choices.<br />
In Eq. (9), we have defined the government’s inflation target as ˆπ. The<br />
fact that the same inflation target appears in Eq. (10) would be correct for<br />
the cases where the bank has the instrument independence, but not target<br />
independence. However, it is easy to relax this assumption and allow the<br />
central bank to choose its own target, as the ECB does. But, as I show in<br />
Hughes Hallett (2005), there is no advantage in doing so since the government<br />
would simply adjust its parameters to compensate. Hence, only if the<br />
bank is free to choose the value of λ cb as well, do we get an extra advantage.<br />
Yet, even this will not be enough to outweigh the advantages to be gained<br />
from fiscal leadership — for the reasons noted in Section 5.<br />
A second feature is that Eqs. (9) and (10) specify symmetric inflation<br />
targets around ˆπ. Symmetric inflation targets are particularly emphasized<br />
as being required of both monetary policy and the fiscal authorities (HMT,<br />
2003). We have, therefore, specified both to be features of the government<br />
and the central bank objective functions here. However, it is not clear that<br />
symmetry has been a feature of the European policy making in practice.<br />
Indeed, the ECB has acquired a reputation for being more concerned to<br />
tighten monetary policy when π> ˆπ, than it is to loosen it when π< ˆπ.<br />
Consequently, rather than symmetry in the output-gap target, we have<br />
asymmetric penalties, which tighten the fiscal constraints in recessions,<br />
but weaken them in a boom when the debt and deficits would be falling<br />
anyway. On one hand, this might not be ideal, since it introduces an element<br />
of pro-cyclicality; on the other hand, it brings the rule closer to reality.
86 Andrew Hughes Hallet<br />
5.4. Institutional Design and Policy Choices<br />
We characterize the strategic interaction between the government and the<br />
central bank as a two-stage non-co-operative game in which the structure of<br />
the model and the objective functions are common knowledge. In the first<br />
stage, the government chooses the institutional parameters δ and λ cb . The<br />
second stage is a Stackelberg game in which fiscal policy takes on a leadership<br />
role. In this stage, the government and the monetary authority set their<br />
policy instruments, given the δ and λ cb values determined at the previous<br />
stage. Private agents understand the game and form rational expectations<br />
for future prices in the second stage. Formally, the policy game runs as<br />
follows:<br />
5.4.1. Stage 1<br />
The government solves the problem:<br />
{ 1<br />
min ELg (g t, m t, δ, λ cb ) = E<br />
δ,λ cb 2 [π t(g t ,m t ) −ˆπ] 2 − λ 2 1 [y t(g t ,m t )]}<br />
+ λg 2<br />
2 E[(b − θ)y t(g t ,m t ) − τ t (g t ,m t )] 2 (11)<br />
where L g (g t ,m t ,δ,λ cb ) is Eq. (9) evaluated at (g t ,m t ,δ,λ cb ), and E<br />
denotes expectations.<br />
5.4.2. Stage 2<br />
(a) Private agents form rational expectations about future prices πt e, before<br />
the shocks u t and ε t are realized.<br />
(b) The shocks u t and ε t are realized and observed by both the government<br />
and the central bank.<br />
(c) The government chooses g t , before m t is chosen by the central bank,<br />
to minimize L g (g t ,m t , ¯δ, ¯λ cb ) where ¯δ and ¯λ cb are at the values determined<br />
at stage 1.<br />
(d) The central bank then chooses m t , taking g t as given, to minimize:<br />
L cb (g t ,m t , ¯δ, ¯λ cb ) = (1 − ¯δ)<br />
[π t (g t ,m t ) −ˆπ] 2<br />
2<br />
− (1 − ¯δ)¯λ cb [y t (g t ,m t )] + ¯δL g (g t ,m t , ¯δ, ¯λ cb )<br />
(12)
The Advantages of Fiscal Leadership in an Economy 87<br />
We solve this game by solving the second stage (for the policy choices)<br />
first, and then substituting the results back into Eq. (11) to determine the<br />
optimal operating parameters δ and λ cb . From stage 2, we get:<br />
π t (δ, λ cb ) =ˆπ + (1 − δ)β(φ − η)λcb + δ(βφ + γ)λ g 1<br />
α[β(φ − η) + δ(βη + γ)]<br />
(13)<br />
y t (δ, λ cb ) =−u t /α (14)<br />
where<br />
τ t (δ, λ cb ) = (1 − δ)βs(βη + γ)(λcb − λ g 1 )<br />
[β(φ − η) + δ(βη + γ)]λ 9 − (b − θ)u t<br />
α<br />
2<br />
η = ∂m t<br />
∂g t<br />
= −α2 γβs 2 + δφλ g 2<br />
(αβs) 2 + δ 2 λ g 2<br />
(15)<br />
(16)<br />
φ = 1 + αβ − γθs (17)<br />
and<br />
= 1 + αβ + βθs. (18)<br />
Evidently, is positive. We assume φ to be positive as well. One would<br />
certainly expect φ>0 since, with θ0.<br />
Nevertheless, the conflict which stands revealed within the fiscal policy is<br />
important. In order to get φ0.
88 Andrew Hughes Hallet<br />
Substituting Eqs. (13)–(15) back into Eq. (11), we can now get the<br />
stage 1 solution from:<br />
{<br />
min ELg (δ, λ cb ) = 1 (1 − δ)β(φ − η)λ cb + δ(βφ + γ)λ g } 2<br />
1<br />
δ,λ cb 2 α[β(φ − η) + δ(βη + γ)]<br />
{<br />
+ λg 2 (1 − δ)βs(βη + γ)(λ cb − λ g 1 )<br />
} 2<br />
2 [β(φ − η) + δ(βη + γ)]λ g . (19)<br />
2<br />
This part of the problem has first-order conditions:<br />
and<br />
(1 − δ)(φ − η)λ g 2 {(1 − δ)β(φ − η)λcb + δ(βφ + γ)λ g 1 }<br />
−(1 − δ) 2 (βη + γ) 2 α 2 s 2 β(λ g 1 − λcb ) = 0 (20)<br />
{(1 − δ)β(φ − η)λ cb + δ(βφ + γ)λ g 1 }<br />
× (λ g 1 − λcb ) {δ(1 − δ) + (φ − η)} λ g 2<br />
−(1 − δ)(βη + γ)α 2 s 2 β {(βη + γ) − (1 − δ)β} (λ g 1 − λcb ) 2 = 0.<br />
(21)<br />
where = ∂η/∂δ. There are two real-valued solutions which satisfy this<br />
pair of first-order conditions. 19 Both are satisfied when δ = 1 and λ cb = λ g 1 .<br />
This solution describes a fully dependent central bank, which is not appropriate<br />
in the Eurozone case. And, it turns out to be inferior to the second<br />
solution: δ = λ cb = 0. In this solution, the central bank is fully independent<br />
and exclusively concerned with the economy’s inflation performance.<br />
Out of the two possibilities, the solution which yields the lowest welfare<br />
loss, as measured by the government’s (society’s) loss function, can be<br />
identified by comparing Eq. (19) to the expected loss that would be suffered<br />
under the alternative institutional arrangement. Substituting, δ = 1 and<br />
λ cb = λ g 1 in Eq. (19) results in: EL g = (λg 1 )2<br />
2α 2 . (22)<br />
19 Because η is a function of δ, Eq. (21) is quartic in δ. This polynomial has four distinct<br />
roots, of which only two are real-valued. The complete solution may be found in Hughes<br />
Hallett and Weymark (2002).
The Advantages of Fiscal Leadership in an Economy 89<br />
But substituting, δ = λ cb = 0 in the right-hand-side of Eq. (19) yields:<br />
EL g = 0. (23)<br />
Consequently, our results show that, when there is fiscal leadership, society’s<br />
welfare loss (as measured by Eq. (19)) is minimized when the government<br />
appoints independent central bankers who are concerned only with<br />
the achievement of a mandated inflation target and completely disregard<br />
the impact their policies may have on the output.<br />
However, our results also indicate that fiscal leadership with an independent<br />
central bank can be beneficial under more general conditions. When<br />
δ = 0, βη + γ = 0 and the externalities between policy makers are neutralized.<br />
20 As a result, Eq. (19) will become:<br />
{ λ<br />
cb<br />
} 2<br />
(24)<br />
EL g = 1 2<br />
α<br />
for any value of λ cb . Hence, an independent central bank will always produce<br />
better results than a dependent one, so long as it is more conservative<br />
than the government (λ cb
90 Andrew Hughes Hallet<br />
This is always smaller than the loss incurred when fiscal leadership is combined<br />
with a dependent central bank. However, the optimal degree of conservatism<br />
for an independent central bank, in this case, is obtained by setting<br />
δ = 0 in Eq. (25) to yield:<br />
λ cb∗ = (αγs)2 λ g 1<br />
(αγs) 2 + φ 2 λ g . (27)<br />
2<br />
It is straightforward to show that the value of EL g in Eq. (24) is always<br />
less than Eq. (26) as long as:<br />
λ cb < [λ g 1 λcb∗ ] 1/2 . (28)<br />
It is also evident that λ cb∗ 0. Consequently, fiscal leadership<br />
with any λ cb
The Advantages of Fiscal Leadership in an Economy 91<br />
Comparing these two sets of outcomes, we have five conclusions. These<br />
conclusions signify the main results of this paper:<br />
(a) Fiscal leadership eliminates the inflationary bias, and results in lower<br />
inflation without any loss in output or output volatility. The proximate<br />
cause of this surprising result is that optimization under fiscal leadership<br />
leads to higher taxes and larger debt repayments. 22<br />
(b) The deeper reason is that there is a self-limiting aspect to the longrun<br />
design of fiscal policy. Unless the effect of fiscal policy on output<br />
is so large as to generate savings/taxes that could finance both fiscal<br />
expansions and debt re-payments (this possibility is ruled out in the<br />
deficit rule case), each expansion designed to increase output would<br />
simply be accompanied by a greater burden of debt. Hence, in order to<br />
preserve the long-run sustainability of its finances, the government must<br />
eventually raise taxes. This fact takes some inflationary pressure off the<br />
central bank when fiscal expansions are called for. As a result, the bank<br />
is less likely to tighten the monetary policy, and externalities are reduced<br />
on both sides. This is what generates a better co-ordination between<br />
the policy makers, as noted in Section 5. But, this can only happen<br />
if the government has a genuine commitment to the long-run goals<br />
(sustainable public finances, and a debt or deficit rule in this case), since<br />
only then will the fiscal policy leader take into account the predictable<br />
reactions (of the follower) to any short-run deviations by the leader.<br />
The follower knows that the leader is not likely to sacrifice his long-run<br />
goals by making such deviations (and anyway has the opportunity to<br />
clear up the mess, according to the follower’s own preferences, if the<br />
leader does so). Thus, in the short-run under simultaneous moves, each<br />
player fails to consider the predictable response, and costs, which the<br />
externalities he/she imposes on his/her rival would create. The ability to<br />
co-ordinate would be lost, and the outcomes worse for both. We show<br />
this in Section 5 (and in a more formal analysis in Hughes Hallett,<br />
2005).<br />
(c) These results hold independently of the commitment to the debt rule<br />
(λ g 2<br />
), or its target value (θ), and independently of the government’s preference<br />
to stabilize or spend (λ g 1<br />
,s), and of the economy’s transmission<br />
parameters (α,β,γ)or the particular value of b as discussed above.<br />
22 Taxes are lower under simultaneous moves because λ cb
92 Andrew Hughes Hallet<br />
(d) Since Eτt ∗ < 0 in Eq. (32), but Eτ t = 0 in (29), the simultaneous moves<br />
regime will always end up in increasing the deficit levels. Therefore, it<br />
will eventually exceed any limit set for the debt ratio. Fiscal leadership<br />
will do neither of these things. More generally, the simultaneous moves<br />
game always leads to fiscal expansions if money financing is small. 23<br />
Indeed, the expected value of s(byt<br />
∗ − τt ∗ ) in Eq. (5) is zero under<br />
fiscal leadership, but φλ cb∗ /(α 2 γ) > 0 in the simultaneous moves case<br />
(as expected, since the monetary policy externalities would be larger).<br />
And since revenues are lower in the latter case, budget deficits are<br />
larger. Hence, the gains from the fiscal leadership are more than just<br />
better outcomes. Leadership also implies less expansionary budgets<br />
and tighter public expenditure controls. The ECB should find this a<br />
significantly more comfortable environment to operate in.<br />
(e) The simultaneous moves regime approaches fiscal leadership, as the<br />
debt rule becomes progressively “harder”: that is, EL g → 0,λ cb∗ →<br />
0; πt ∗ →ˆπ, and Eτt ∗ → 0, as λ g 2 →∞.<br />
6. The Co-Ordination Effect<br />
In our model, the central bank is independent. Without further institutional<br />
restraints, interactions between an independent central bank and the government<br />
would lead to non-co-operative outcomes. But, if the government<br />
is committed to long-term leadership in the manner that I have described,<br />
the policy game will become an example of rule-based co-ordination in<br />
which both parties can gain without any reduction in the central bank’s<br />
independence. However, this is not to say that both parties gain equally. If<br />
the inflation target is strengthened after the fiscal policy is set, the leadership<br />
gains to the government will be less (Hughes Hallett and Viegi, 2002). This<br />
is an important point because it implies that granting leadership to a central<br />
bank whose inflation aversion is already greater than the government’s, or<br />
whose commitment is greater, will produce no additional gains. For this<br />
reason, granting leadership to a central bank that cannot influence debt or<br />
deficit directly, but puts a higher priority on inflation control will bring<br />
smaller gains from a society’s point of view (whatever it may do for the<br />
central bank). I demonstrate these points formally in Hughes Hallett (2005).<br />
23 m t ≈ 0; see Appendix B. This result still holds good when β>γ, even if monetary<br />
financing is allowed. It also holds good if the central bank is conservative, or the government<br />
is liberal, when β
The Advantages of Fiscal Leadership in an Economy 93<br />
Finally, given the structure of the Stackelberg game, there is no incentive<br />
to re-optimize for either party — so long as the government remains<br />
committed to long-term leadership rather than short-term demand management<br />
— unless both parties agree to reduce their independence through<br />
discretionary co-ordination. 24 This is a general result: it holds good for any<br />
model where both inflation and output depend on both monetary and fiscal<br />
policy, and where inflation is targeted to some degree by both players<br />
(Demertzis et al., 2004; Hughes Hallett and Viegi, 2002). Thus, although<br />
Pareto improvements, over no co-operation, do not emerge for both parties<br />
in all Stackelberg games, they do so in problems of the kind examined here.<br />
Our results are robust.<br />
6.1. Empirical Evidence<br />
Whether or not these results are of practical importance is another matter.<br />
In Table 3, I have computed the expected losses under the simultaneous<br />
move and fiscal leadership regimes for 10 countries when both fiscal policy<br />
and the central bank are constructed optimally. The data I have used is from<br />
1998, the year in which the Eurozone was created. The data itself, and its<br />
sources, are summarized in Appendix A.<br />
The countries that are selected fall into three groups:<br />
(a) Eurozone countries, with independently set fiscal and monetary policies<br />
and target independence: France, Germany, Italy, and the Netherlands;<br />
(b) Explicit inflation targeters with fiscal leadership: Sweden, New<br />
Zealand, and the United Kingdom and<br />
(c) Federalists, with joint inflation and stabilization objectives: United<br />
States, Canada, and Switzerland.<br />
In the first group, monetary policy is conducted at the European level<br />
and fiscal policy at the national level. Policy interactions, in this group,<br />
can be characterized in terms of a simultaneous move game, with target as<br />
well as with instrument independence for both sets of policy authorities.<br />
24 This supplies the political economy of our results. We need no pre-commitment beyond<br />
leadership (the Stackelberg game supplies sub-game perfection, see Note 4) and no punishment<br />
beyond electoral results. The former will hold so long as governments have a natural<br />
commitment to maintaining sustainable public finances, public services (health, education,<br />
and defence), or social equity, all long-run “contractual” issues. The latter will then arise<br />
from the political competition inherent in any democracy.
94 Andrew Hughes Hallet<br />
Table 3.<br />
Expected losses under a deficit rule, leadership vs. other strategies.<br />
Full Fiscal Simultaneous Losses under<br />
dependence leadership moves simultaneous<br />
δ = 1 δ = 0; λ g 1 = 1 1 = 1 moves: Growth-<br />
λ cb = λ g 1 λ cb = 0<br />
λ cb = λ cb∗ rate equivalents<br />
France 5.78 0.00 0.134 13.4<br />
Germany 16.14 0.00 0.053 5.30<br />
Italy 1.28 0.00 0.207 20.7<br />
Netherlands 1.28 0.00 0.165 16.5<br />
Sweden 4.51 0.00 0.034 3.40<br />
New Zealand 8.40 0.00 0.071 7.10<br />
United Kingdom 3.37 0.00 0.057 5.70<br />
United States 6.46 0.00 0.248 24.8<br />
of America<br />
Canada 12.50 0.00 0.118 11.8<br />
Switzerland 4.79 0.00 0.066 6.60<br />
It is, perhaps, arguable that the Eurozone has adopted a monetary leadership<br />
regime; but the empirical evidence for this was weak, and the fiscal<br />
constraints that could have sustained such a leadership were widely ignored.<br />
The second group of countries has adopted explicit, and mostly publicly<br />
announced, inflation targets. Central banks in these countries have<br />
been granted instrument independence, but not target independence. The<br />
government either sets, or helps to set, the inflation target. In each case, the<br />
government has adopted long-term (supply side) fiscal policies, leaving an<br />
active demand management to monetary policy. These are clear cases in<br />
which there is both fiscal leadership and instrument independence for the<br />
central bank.<br />
The third group represents a set of more flexible economies, with<br />
implicit inflation targeting and a statutory concern for stabilization; also federalism<br />
in fiscal policy and independence at the central bank, and therefore<br />
no declared leadership in either fiscal or monetary policies. This provides<br />
a useful benchmark for the other two groups.<br />
The performance under the different fiscal regimes when the fiscal constraint<br />
is a relatively soft deficit rule is reported in Table 3. Column 1<br />
shows the losses under a dependent central bank in welfare units — leadership<br />
or not. Column 2 reflects the losses that would be incurred under
The Advantages of Fiscal Leadership in an Economy 95<br />
the government leadership with an independent central bank that directs<br />
monetary policy exclusively towards the achievement of the inflation target<br />
(i.e., δ = λ cb = 0). The third column gives the minimum loss associated<br />
with a simultaneous decision-making version of the same game. In each<br />
case, the deficit rule is soft for the Eurozone countries (λ g 2<br />
= 0.25); medium<br />
strength for the United States, Canada, and Switzerland (λ g 2<br />
= 0.5); and<br />
harder for Sweden, the United Kingdom and New Zealand (λ g 2<br />
= 1). The<br />
deficit ratio targets are 0% in all countries (a balanced budget in the medium<br />
term), but the propensity to spend, s, out of any remaining deficit or any<br />
endogenous budget improvements is one.<br />
Evidently, complete dependence in monetary policy is extremely unfavorable,<br />
although the magnitude of the loss varies considerably from country<br />
to country. The losses in Column 3 (Table 3) appear to be relatively small<br />
in comparison. However, when these figures are converted into “growth-rate<br />
equivalents”, I find these losses to be significant. A growth-rate equivalent<br />
is the loss in output growth that would produce the same welfare loss, if all<br />
other variables remain fixed at their optimized values. 25<br />
The figures in Column 4 are more interesting. They show that the losses<br />
associated with simultaneous decision-making are equivalent to having permanent<br />
reductions of up to 20% in the level of national income. That is,<br />
France, Italy, and the Netherlands could have expected to have grown about<br />
15%–20% more, had they adopted fiscal leadership with deficit targets; and<br />
Sweden, the United Kingdom, and New Zealand around 3%–5% less, had<br />
they not done so. Similarly, fiscal leadership with debt targeting would<br />
be worth 10%–20% extra GDP in the United States, and Canada. These<br />
are significant gains and significantly larger than could be expected from<br />
international policy co-ordination itself (Currie et al., 1989).<br />
Thus, the immediate effect of introducing a stability pact deficit rule<br />
has been to increase the costs of an unco-ordinated (simultaneous moves)<br />
regime dramatically. This reveals the weakness of the deficit rule as a<br />
25 Currie et al. (1989). To obtain these figures, I compute the marginal rates of transformation<br />
around each government’s indifference curve to find the change in output growth, dy t , that<br />
would yield the welfare losses in Column 3. Formally,<br />
dy t =<br />
dEL g t<br />
[λ g 2 {(b − θ)y t − τ t }(b − θ) − λ g 1 ]<br />
using Eq. (9). The minimum value of dy t is, therefore, attained when tax revenues τ t grow at<br />
the same rate as the re-payment target (b − θ)y t . These are the losses reported in Column 4.
96 Andrew Hughes Hallet<br />
restraint mechanism. Because it is a flow and not a stock, it has no inherent<br />
persistence. Moreover, a larger part (the current automatic stabilizers,<br />
social expenditures, and tax revenues) will be endogenous, varying with the<br />
level of economic activity and with the pressures of the electoral cycle. It is,<br />
therefore, much harder to use as a commitment device with any credibility<br />
in the long term. If we assume that it can be pre-committed, as in the leadership<br />
scenario, then we get good results of course. But, if this assumption<br />
is likely to be violated (or challenged), then the fact that the results are so<br />
much worse than in the simultaneous moves case shows how difficult it<br />
is to pre-commit deficits in advance. Being only a small component in a<br />
moving total, it does not have the persistence of a cumulated debt criterion<br />
because violations are not carried forward. So, when policy conflicts arise,<br />
the deficit either fails its target or it has to be restrained to such a degree<br />
that it starts to do serious damage to overall economic performance.<br />
7. Conclusions<br />
I conclude this chapter with the following observations:<br />
(a) Fiscal leadership leads to improved outcomes because it implies a<br />
degree of co-ordination and reduced conflicts between institutions,<br />
without the central bank having to lose its ability to act independently.<br />
This places the outcomes somewhere between the superior (but discretionary)<br />
policies of complete co-operation; and the non-co-operative<br />
(or rule-based) policies of complete independence.<br />
(b) The co-ordination gains come from the self-limiting action of fiscal<br />
policy when fiscal policy is given a long-run focus in the form of a<br />
(soft) debt or deficit rule. Leadership is the crucial element here; these<br />
gains do not appear in a straight-forward competitive regime with the<br />
same debt or deficit rule.<br />
(c) The leadership model predicts improvements in inflation and the fiscal<br />
targets, without any loss in growth or output volatility. In addition,<br />
leadership requires less precision in the setting of the strategic or institutional<br />
parameters. It is easier to implement.<br />
(d) Debt ratios will increase in a policy game without co-ordination, but<br />
do not do so under fiscal leadership. Likewise, deficit rules without<br />
co-ordination to restrain fiscal policy imply larger deficits (and more<br />
expansionary policies) than do debt rules of a similar specification<br />
because deficits (being a flow not a stock, and with less natural persistence)<br />
are less easy to pre-commit credibly. These two results explain
The Advantages of Fiscal Leadership in an Economy 97<br />
why the ECB prefers hard restraints (and why the fiscal policy makers<br />
do not), and suggests that the ECB may actually prefer fiscal leadership<br />
because of the greater degree of pre-commitment implied.<br />
References<br />
Barro, RJ (1981). Output effects of government purchases. Journal of Political Economy<br />
89, 1086–1121.<br />
Barro, RJ and DB Gordon (1983). Rules, discretion, and reputation in a model of monetary<br />
policy. Journal of Monetary <strong>Economic</strong>s 12, 101–121.<br />
Basar, T (1989). Time consistency and robustness of equilibria in non-cooperative dynamic<br />
games. In Dynamic Policy Games in <strong>Economic</strong>s, F van der Ploeg and A de Zeeuw<br />
(eds.), pp. 9–54. Amsterdam: North Holland.<br />
Basar, T and G Olsder (1989). Dynamic noncooperative game theory. In Classics in Applied<br />
Mathematics, SIAM series, Philadelphia: SIAM.<br />
Brandsma, A and A Hughes Hallett (1984). <strong>Economic</strong> conflict and the solution of dynamic<br />
games. European <strong>Economic</strong> Review 26, 13–32.<br />
Buti, M, S Eijffinger and D Franco (2003). Revisiting the stability and growth pact: grand<br />
design or internal adjustment? Discussion Paper 3692. London: Centre for <strong>Economic</strong><br />
Policy Research.<br />
Canzoneri, M (2004). A New View on The Transatlantic Transmission of Fiscal Policy<br />
and Macroeconomic Policy Coordination. In The Transatlantic Transmission of Fiscal<br />
Policy and Macroeconomic Policy Coordination, M Buti (ed.), Cambridge: Cambridge<br />
University Press.<br />
Currie, DA, G Holtham andA Hughes Hallett (1989). The theory and practice of international<br />
economic policy coordination: does coordination pay? In Macroeconomic Policies in<br />
an Interdependent World, R Bryant, D Currie, J Frenkel, P Masson and R Portes (eds.),<br />
Washington, DC: International Monetary Fund, 125–148.<br />
Demertzis, M, A Hughes Hallett and N Yiegi (2004). An Independent Central Bank Faced<br />
with Elected Governments: A Political Economy Conflict. In European Journal of<br />
Political Economy 20(4), 907–922.<br />
Dieppe, A, K Kuster and P McAdam (2004). Optimal monetary policy rules for the Euro<br />
Area: an analysis using the area wide model. Working Paper 360. Frankfurt: European<br />
Central Bank.<br />
Dixit,A (2001). Games of monetary and fiscal interactions in the EMU. European <strong>Economic</strong><br />
Review 45, 589–613.<br />
Dixit, AK and L Lambertini (2003). Interactions of commitment and discretion in monetary<br />
and fiscal issues. American <strong>Economic</strong> Review 93, 1522–1542.<br />
Dow, JCR (1964). The Management of the British Economy 1945–1960. Cambridge:<br />
Cambridge University Press.<br />
European Commission (2002). Public finances in EMU:2002. European Economy: Studies<br />
and Reports 4, Brussels: European Commission.<br />
Fatas, A, J von Hagen, A Hughes Hallett, R Strauch and A Sibert (2003). Stability and<br />
Growth in Europe. London: Centre for <strong>Economic</strong> Policy Research (MEI-13).<br />
Gali, J and R Perotti (2003). Fiscal policy and monetary integration in Europe. <strong>Economic</strong><br />
Policy 37, 533–571.
98 Andrew Hughes Hallet<br />
HM Treasury (HMT) (2003). Fiscal Stabilization and EMU. HM Stationary Office, Cmnd<br />
799373. Norwich: also available from www.hm-treasury.gov.uk.<br />
Hughes Hallett, A (1984). Noncooperative strategies for dynamic policy games and the<br />
problem of time inconsistency. Oxford <strong>Economic</strong> Papers 36, 381–399.<br />
Hughes Hallett,A (2005). In praise of fiscal restraint and debt rules: what the Euro zone might<br />
do now. Discussion Paper 5043. London: Centre for <strong>Economic</strong> Research, 251–279.<br />
Hughes Hallett, A and D Weymark (2002). Government leadership and central bank design.<br />
Discussion Paper 3395, London: Centre for <strong>Economic</strong> Research.<br />
Hughes Hallett, A and N Viegi (2002). Inflation targeting as a coordination device. Open<br />
Economies Review 13, 341–362.<br />
Hughes Hallett, A and D Weymark (2004a). Policy games and the optimal design of central<br />
banks. In Money Matters, P Minford (ed.). London: Edward Elgar.<br />
Hughes Hallett,A and DWeymark (2004b). Independent monetary policies and social equity.<br />
<strong>Economic</strong>s Letters 85, 103–110.<br />
Hughes Hallett,A and D Weymark (2005). Independence before conservatism: transparency,<br />
politics and central bank design. German <strong>Economic</strong> Review 6, 1–25.<br />
Lucas, RE (1972). Expectations and the neutrality of money. Journal of <strong>Economic</strong> Theory<br />
4, 103–124.<br />
Lucas, RE (1973). Some international evidence on output-inflation trade-offs. American<br />
<strong>Economic</strong> Review 63, 326–334.<br />
McCallum, BT (1989). Monetary <strong>Economic</strong>s: Theory and Policy. New York: Macmillan.<br />
Svensson, L (2003). What is wrong with Taylor rules? Using judgement in monetary rules<br />
through targeting rules. Journal of <strong>Economic</strong> Literature 41, 426–477.<br />
Taylor, JB (1993a). Macroeconomic Policy in a World Economy: From Econometric Design<br />
to Practical Operation. New York: W.W. Norton and Company.<br />
Taylor, JB (1993b). Discretion vs. policy rules in practice. Carnegie-Rochester Conference<br />
Series on Public Policy 39, 195–214.<br />
Taylor, JB (2000). Discretionary fiscal policies. Journal of <strong>Economic</strong> Perspectives 14, 1–23.<br />
Turrini, A and J in ‘t Veld (2004). The Impact of the EU Fiscal Framework on <strong>Economic</strong><br />
Activity. DGII, Brussels: European Commission.<br />
Appendix A: Data Sources for Table 3<br />
The data used in Table 3 are set out in the table below. They come from different<br />
sources and represent “stylized facts” for the corresponding parameters.<br />
The Phillips curve parameter, α, is the inverse of the annualized<br />
sacrifice ratios on quarterly data from 1971–1998. The values for β and γ,<br />
the impact multipliers for fiscal and monetary policy, respectively, are the<br />
one-year simulated multipliers for these policies in Taylor’s multicountry<br />
model (Taylor, 1993a). I have calibrated the remaining parameters using<br />
stylized facts from 1998: the year that EMU started, and the year that new<br />
fiscal and monetary regimes took effect in the United Kingdom and Sweden.<br />
Thus, s is the automatic stablizer effect on output according to the European<br />
Commission (2002) and HMT (2003) estimates, allowing for larger social
The Advantages of Fiscal Leadership in an Economy 99<br />
Table 4.<br />
Data for the deficit rule calculations.<br />
α β γ s θ φ λ g 2<br />
France 0.294 0.500 0.570 1 0 1.147 0.25<br />
Germany 0.176 0.533 0.430 1 0 1.094 0.25<br />
Italy 0.625 0.433 0.600 1 0 1.271 0.25<br />
Netherlands 0.625 0.489 0.533 1 0 1.306 0.25<br />
Sweden 0.333 0.489 0.533 1 0 1.163 1<br />
New Zealand 0.244 0.400 0.850 1 0 1.098 1<br />
UK 0.385 0.133 0.580 1 0 1.038 1<br />
United States 0.278 0.467 1.150 1 0 1.130 0.5<br />
Canada 0.200 0.400 0.850 1 0 1.080 0.5<br />
Switzerland 0.323 0.489 0.533 1 0 1.158 0.5<br />
security sectors in the Netherlands and Sweden. Similarly, θ is set to give a<br />
debt target somewhat below each country’s 1998 debt ratio and the SGP’s<br />
60% limit. Likewise, λ g 2<br />
was set to imply a low, high, or medium commitment<br />
to debt/deficit limits, as reflected in actual behavior in 2004. Finally,<br />
I have set λ g 1<br />
= 1 in each country to suggest that governments, if left to<br />
themselves, are equally concerned about inflation and output stabilization.<br />
For Table 4, I suppose that the deficit that remains after growth effects<br />
and new discretionary taxes are accounted for will be spent (s = 1), and<br />
that the medium-term deficit target is zero (θ = 0).<br />
Appendix B: Derivation of the Change in Fiscal Stance<br />
Between Regimes<br />
The expected value of s(by t − τ t ) under fiscal leadership and simultaneous<br />
moves can be obtained by inserting values first from Eq. (29), and then from<br />
Eqs. (31) and (32), respectively. Taking the expected values, this yields<br />
zero and φλ cb∗ /(α 2 γ) > 0 in each case. Applying this result to Eq. (4),<br />
it shows that the expected change in output between regimes, which we<br />
know to be zero, is (β − γ)m − βλ cb∗ /(αλ g 1<br />
) + γ[s(by − τ)] = 0.<br />
Hence,<br />
m ={γ[s(by − τ)] + βλ cb∗ /(αλ g 1<br />
)}/(β − γ)<br />
which, given the change in s(by t − τ t ), is positive, if β>γ. In this case,<br />
the simultaneous move regime is unambiguously more expansionary, and<br />
will run larger deficits, than a regime with fiscal leadership. But, since
100 Andrew Hughes Hallet<br />
γ/(β − γ) > 1, this result may not follow if β
Topic 2<br />
Cheap-Talk Multiple Equilibria and Pareto —<br />
Improvement in an Environmental<br />
Taxation Games<br />
Chirstophe Deissenberg<br />
UniversitédelaMéditerranée and GREQAM, France<br />
Pavel Ševčík<br />
Universilé de Montréal, Canada<br />
1. Introduction<br />
The use of taxes as regulatory instrument in environmental economics is a<br />
classic topic. In a nutshell, the need for regulation usually arises because<br />
producing causes detrimental emissions. Due to the lack of a proper market,<br />
the firms do not internalize the impact of these emissions on the utility of<br />
other agents. Thus, they take their decisions on the basis of prices that do not<br />
reflect the true social costs of their production. Taxes can be used to modify<br />
the prices, confronting the firms so that the socially desirable decisions are<br />
taken. The problem has been exhaustively investigated in static settings,<br />
where there is no room for strategic interaction between the regulator and<br />
the firms.<br />
Consider, however, the following situations:<br />
(1) The emission taxes have a dual effect, they incite the firms to reduce<br />
production and to undertake investments in abatement technology. This<br />
is typically the case, when the emissions are increasing in the output<br />
and decreasing in the abatement technology;<br />
(2) Emission reduction is socially desirable. The reduction of production<br />
is not and<br />
(3) The investments are irreversible. In this case, the regulator must find<br />
an optimal compromise between implementing high taxes to motivate<br />
high investments, and keeping the taxes low to encourage production.<br />
101
102 Chirstophe Deissenberg and Pavel Ševčík<br />
The fact that the investments are irreversible introduces a strategic<br />
element in the problem. If the firms are naïve and believe his announcements,<br />
the regulatory can insure high production and important investments<br />
by first declaring high taxes and reducing them, once the corresponding<br />
investments have been realized. More sophisticated firms, however, recognize<br />
that the initially high taxes will not be implemented, and are reluctant<br />
to invest in the first place. In other words, one is confronted with a typical<br />
time inconsistency problem, which has been extensively treated in the monetary<br />
policy literature following Kydland and Prescott (1977) and Barro and<br />
Gordon (1983). In environmental economics, the time inconsistency problem<br />
has received yet only limited attention, although it frequently occurs.<br />
See among others Gersbach and Glazer (1999) for a number of examples<br />
and for an interesting model, see Abrego and Perroni (1999); Batabyal<br />
(1996a;b); Dijkstra (2002); Marsiliani and Renstrom (2000) and Petrakis<br />
and Xepapadeas (2003).<br />
The time inconsistency is directly related to the fact that the situation<br />
described above, defines a Stackelberg game between the regulator (the<br />
leader) and the firms (the followers). As noted in the seminal work of<br />
Simaan and Cruz (1973a;b), inconsistency arises because the Stackelberg<br />
equilibrium is not defined by mutual best responses. It implies that the<br />
follower uses a best response in reaction to the leader’s action, but not that<br />
the leader’s actions is itself a best response to the follower’s. This opens<br />
the door to a re-optimization by the leader once the follower has played.<br />
Thus, a regulator, who announces that it will implement the Stackelberg<br />
solution, is not credible. An usual conclusion is that, in the absence of<br />
additional mechanisms, the economy is doomed to converge towards the<br />
less desirable Nash solution.<br />
A number of options to insure credible solutions have been considered<br />
in the literature — credible binding commitments by the Stackelberg<br />
leader, reputation building, use of trigger strategies by the followers, etc.<br />
(see McCallum (1997), for a review in a monetary policy context). Schematically,<br />
these solutions aim at assuring the time consistency of Stackelberg<br />
solution with either the regulator or the firms as a leader. Usually, these<br />
solutions are not efficient and can be Pareto-improved.<br />
In Dawid et al. (2005), we demonstrated the usefulness of a new solution<br />
to the time inconsistency problem in environmental policy. We showed<br />
that tax announcements, that are not respected, can increase the payoff not<br />
only of the regulator, but also of all firms, if these include any number<br />
of naïve believers who take the announcements at face value. Moreover,<br />
if firms tend to adopt the behavior of the most successful ones, a unique
Cheap-talk Multiple Equilibria and Pareto 103<br />
stable equilibrium may exist where a positive fraction of firms are believers.<br />
This equilibrium Pareto-dominates the one, where all firms anticipate<br />
perfectly the Regulator’s action. To attain the superior equilibrium, the<br />
regulator builds reputation and leadership by making announcements and<br />
implementing taxes in a way that generates good results for the believers,<br />
rather than by pre-committing to his announcements.<br />
The potential usefulness of employing misleading announcements to<br />
Pareto-improve, upon standard game-theoretic equilibrium solutions was<br />
suggested for the case of general linear-quadratic dynamic games in Vallee<br />
et al. (1999) and developed by the same authors in subsequent papers.<br />
An early application to environmental economies is Vallee (1998). The<br />
believers/non-believers dichotomy was introduced by Deissenberg and<br />
Gonzalez (2002), who study the credibility problem in monetary economics<br />
in a discrete-time framework with reinforcement learning. A similar monetary<br />
policy problem has been investigated by Dawid and Deissenberg (2005)<br />
in a continuous-time setting akin to the used in the present work.<br />
In this chapter, we re-visit the model proposed by Dawid et al. (2005)<br />
and show that a minor (and plausible) modification of the regulator’s objective<br />
function, opens the door to the existence of multiple stable equilibria<br />
with distinct basins of attraction — with important interpretations and<br />
potential consequences for the practical conduct of environmental policy.<br />
This chapter is organized as follows. In Section 2, we present the model<br />
of environmental taxation and introduce the imitation-type dynamics that<br />
determinate the evolution of the number to believers in the economy. In<br />
Section 3, we derive and discuss the solution of the sequential Nash game,<br />
one obtains by assuming a constant proportion of believers. In Section 4,<br />
we formulate the dynamic problem and investigate its solution numerically.<br />
Finally, in Section 6, we summarize the mechanisms at work in the model<br />
and the main insights.<br />
2. The Model<br />
We consider an economy, consisting of a Regulator R and a continuum of<br />
atomistic, profit-maximizing firms i with an identical production technology.<br />
Time τ is continuous. To keep the notation simple, we do not index<br />
the variables with either i or r unless it is useful for a better understanding.<br />
In a nutshell, the situation of interest is the following: The regulator<br />
can tax the firms to incite them to reduce their emission, and to generate<br />
tax income, taxes, however, have a negative impact on the employment
104 Chirstophe Deissenberg and Pavel Ševčík<br />
(and thus, on the labor income) and the profits. Thus, R has to choose the<br />
tax level in order to achieve an optimal compromise among tax revenue,<br />
private sector income, emission reduction, and employment levels. The<br />
following sequence of event (S) occurs in every τ:<br />
— R makes a non-binding announcement t a ≥ 0 about the tax level t ≥ 0<br />
it will choose.<br />
—Givent a , the firms form expectations t e about the true level of the<br />
environmental tax. As will be described in more detail later, there are<br />
two different ways for an individual firm to form its expectations.<br />
— Each firm decides about its level of investment v based on its expectation<br />
t e .<br />
— R choose the actual level of tax t ≥ 0. The tax level t is the amount<br />
each firm has to pay per unit of its emissions.<br />
— Each firm produces a quantity x.<br />
— Each firm may revise the way it forms its expectations depending on its<br />
relative profits.<br />
All firms are identical, except for the way they form their expectations t e .<br />
We assume full information: the regulator and the firms perfectly know<br />
both the sequence, we just presented and the economic structure will be<br />
(production and prediction technology, objective functions ...) described<br />
in the remainder of this section.<br />
2.1. The Firms, Bs and NBs<br />
Each firm produces the same homogenous goods using a linear production<br />
technology: the production of x units of output, requires x units of labor and<br />
generates x units of environmentally damaging emissions. The production<br />
costs are given by:<br />
c(x) = wx + c x x 2 , (1)<br />
where x is the output, w>0 the fixed wage rate, and c x > 0 a parameter for<br />
simplicity sake, the demand is assumed infinitely elastic at the given price<br />
˜p >w. Let p := ˜p − w>0. At each point of time, each firm can spend<br />
an additional amount of money γ in order to reduce its current emissions<br />
x. The investment<br />
γ(v) = 1 2 v2 (2)<br />
is needed, in order to reduce the firm’s emissions from x to max [x − v, 0].<br />
The investment in one period has no impact on the emissions in future
Cheap-talk Multiple Equilibria and Pareto 105<br />
period. Rather than expenditures in emission-reducing capital, γ can therefore<br />
be interpreted as the additional costs resulting of a temporary switch<br />
to a cleaner resource — for e.g., of a switch from coal to natural gas.<br />
The firms attempts to maximize their profit. As we shall see at a later<br />
place, one can assume without loss of generality that they disregard the<br />
future and maximize in every τ, their current instantaneous profit. However,<br />
the actual tax rate t enters their instantaneous profit function and is unknown<br />
at the time when they choose v. Thus, they need to base their optimal<br />
choice of v on an expectation (or, prediction) t e of t. In reality, a firm<br />
can invest larger or smaller amounts of money in order to make or less<br />
accurate predictions of the tax, the government will actually implement.<br />
In this model, we assume for simplicity that each firm has two (extreme)<br />
options for predicting the implemented tax. It can either suppose that R’s<br />
announcement is truthful, that is, that t will be equal to t a , or it can do costly<br />
research and try to better predict the t that will actually be implemented.<br />
Depending upon the expectation-formation mechanism the firm chooses,<br />
we speak of believer B or of a non-believer NB:<br />
A believer considers the regulator’s announcement to be truthful<br />
and sets<br />
t e = t a (3)<br />
This does not cost anything.<br />
A non-believer makes a “rational” prediction of t, considering the current<br />
number of Bs and NBs as constant. Making the prediction in any given<br />
period τ costs δ>0.<br />
Details on the derivation of the NBs’prediction will be given later. Note<br />
that a firm can change its choice of expectation-formation mechanism at<br />
any moment in time. That is, a B can become a NB, and vice versa. We<br />
denote with π = π(t) ∈ [0, 1] the current fraction of Bs in the population.<br />
Thus, 1 − π is the current fraction of NBs.<br />
2.2. The Regulator, R<br />
The regulator’s goal is to maximize with respect to t a ≥ 0 and t ≥ 0, the<br />
inter-temporal social welfare function.<br />
(t a ,t)=<br />
∫ ∞<br />
0<br />
e −ρτ ϕ(t a ,t)dτ
106 Chirstophe Deissenberg and Pavel Ševčík<br />
with<br />
ϕ(t a ,t)= (πx B + (1 − π)x NB ) + πg B + (1 − π)g NB<br />
− (π(x B − v B ) + (1 − π)(x NB − v NB )) 2<br />
+ t(π(x B − v B ) + (1 − π)(x NB − v NB )), (4)<br />
where x B ,x NB ,v B ,v NB denote the production respectively, investment<br />
chosen by the believers B and the non-believers NB, and where g B (.)<br />
and g NB (.) are their instantaneous profits,<br />
g B = px B − c x (x B ) 2 − t(x B − v B ) − 1 2 (vB ) 2<br />
g NB = px NB − c x (x NB ) 2 − t(x NB − v NB ) − 1 2 (vNB ) 2 − δ.<br />
The strictly positive parameter ρ is a social discount factor.<br />
Thus, the instantaneous welfare φ is the sum of: (1) the economy’s<br />
output, that is also, the level of employment (remember that output and<br />
employment are one-to-one in this economy); (2) the total profits of Bs<br />
and NBs; (3) the squared volume of emissions and (4) the tax revenue. This<br />
function can be recognized as a reasonable approximation of the economy’s<br />
social surplus. The difference between the results of Dawid et al. (2005) and<br />
those of the present article stem directly from the different specification of<br />
the regulator’s objective function. In the former article, this function does<br />
not include the firm’s profit and is linear in the emissions.<br />
2.3. Dynamics<br />
The firms switch between the two possible expectation-formation mechanisms<br />
(B or NB), according to a imitation-type dynamics, see Dawid (1999);<br />
Hofbauer and Sigmund (1998). Specifically, at each τ the firms meet randomly<br />
two-by-two, each pairing being equiprobable. At every encounter,<br />
the firm with the lower current profit adopts the belief of the other firm with<br />
a probability proportional to the current difference between the individual<br />
profits. Thus, on the average, the firms tend to adopt the type of prediction,<br />
N or NB, which may currently leads to the highest profits. Above<br />
mechanism gives rise to the dynamics.<br />
dπ<br />
=˙π = βπ(1 − π)g B − g NB . (6)<br />
dt<br />
Notice that ˙π reaches its maximum for π =<br />
2 1 (the value or π for<br />
which the probability, for encounter between firms with different profits are<br />
(5)
Cheap-talk Multiple Equilibria and Pareto 107<br />
maximized), and tends towards 0, for π → 0 and π → 1 (for extreme values<br />
of π, all but very few firms have the same profits). The parameter β ≥ 0,<br />
that captures the probability with which a firm changes its strategy during<br />
one of these meetings, calibrates the speed of the information flow between<br />
Bs and NBs. It can be interpreted as a measure of the firms’ willingness to<br />
change strategies, that is, of the flexibility of the firms.<br />
Equation (6) implies that by choosing the value of (t a ,t)at time τ the<br />
regulator not only influences the instantaneous social welfare, but also the<br />
future proportion of Bs in the economy. This, in turn, has an impact on<br />
the future social welfare. Hence, there are no explicit dynamics for the<br />
economy. This again has an impact on the future social welfare. Hence,<br />
although there are no explicit dynamics for the economic variables v and<br />
x, R faces a non-trivial inter-temporal optimization problem.<br />
On the other hand, since the firms are atomistic, each single producer<br />
is too small to influence the dynamics Eq. (6) of π, the only source of<br />
dynamics in the model. Thus, the single firm does not take into account<br />
any inter-temporal effect and, independently of its true planing horizon,<br />
maximizes its current profit in every τ. This naturally leads us to examine<br />
the (Nash) solution of the game between the regulator and the non-believers<br />
that follows from the sequence (S) when, for an arbitrary, fixed value of π,<br />
(a) R maximizes the integrand φ in Eq. (4) with respect to t a and t and (b)<br />
the NBs maximize their profits with respect to ν NB and x NB . As previously<br />
mentioned, we assume full information. In particular, the R and the NBs<br />
are fully aware of the (non-strategic) behavior of the NBs.<br />
The analysis of the static case is also necessary since, as previously<br />
mentioned, it provides the prediction t e of the NBs.<br />
3. The Sequential Nash Game<br />
The game is solved by backwards induction, taking into account the fact<br />
that the firms, being atomistic, rightly assume that their individual actions<br />
do not affect the aggregate values, that is consider these values as given. We<br />
consider only symmetric solutions where all Bs respectively, NBs choose<br />
the same values for v and x.<br />
3.1. Derivation of the Solution<br />
The last decision in the sequence (S) is the choice of the production level x<br />
by the firms. This choice is made after v and t are known. The firms being
108 Chirstophe Deissenberg and Pavel Ševčík<br />
takers, the optimal symmetric production decision is:<br />
x B = x NB = x = p − t<br />
2c x<br />
. (7)<br />
Note that the optimal production is the same for Bs and NBs, as it<br />
exclusively depends upon the realized taxes t.<br />
In the previous stage, R chooses t given v B . Maximizing v NB with<br />
respect to t, with x is given by (7) gives the optimal reaction function:<br />
t(v B ,v NB ) = p − c x[2(πv B + (1 − π)v NB ) + 1]<br />
1 + c x<br />
. (8)<br />
When the firms determine their investment v, Eqs. (7) and (8), and t a<br />
are known, but the actual tax rate t is not. Using Eq. (7) in Eq. (5), one<br />
recognizes that the firms choose v in order to maximize<br />
That is, they set:<br />
(p − t e 2<br />
)<br />
+ t e v − 1 4c x 2 v2 . (9)<br />
v = t e (10)<br />
The Bs predict that t will be equal to its announced value, t e = t a . Thus,<br />
v B = t a . (11)<br />
The NBs know that the regulator will act according to Eq. (8). Thus, they<br />
choose:<br />
v NB = p − c x[2(πv B + (1 − π)v NB ) + 1]<br />
1 + c x<br />
. (12)<br />
Solving this last equation for v NB gives for the equilibrium value:<br />
v NB = v NB (t a ) p − c x(1 + 2πt a )<br />
1 + c x (3 − 2π) . (13)<br />
The first decision to be taken is the choice of t a by R. The regulator’s<br />
instantaneous objective function φ, using the results from the previous<br />
stages, i.e., Eq. (7) for x, Eq. (11) for v B and Eq. (13) for v NB , is concave
with respect to t a for all π ∈ [0, 1], if,<br />
Cheap-talk Multiple Equilibria and Pareto 109<br />
c x > 1/ √ 3. (14)<br />
We shall assume in the following that this inequality holds. Then, the<br />
regulator will choose<br />
t a$ =<br />
1 + p . (15)<br />
1 + 3c x<br />
Using this last equation in Eq. (8) with v NB given by Eq. (13), one obtains<br />
for the equilibrium value of t<br />
t $ =<br />
1 + p 1 + c x<br />
−<br />
1 + 3c x c x (3 − 2π) + 1 . (16)<br />
Both t a$ and t $ will be non-negative if<br />
which we assume from now on.<br />
p ≥<br />
3.2. Some Properties of the Solution<br />
3<br />
3c x − 2 , (17)<br />
For all c x > 0 and π ∈ [0, 1], the optimal announcement t a$ is a constant,<br />
and always less than t a$ . It is optimal to announce a high tax in order to<br />
elicit a high investment from the believers, and to implement a low tax to<br />
motivate a high production level. Moreover, t $ decreases in π: When the<br />
proportion of Bs augments, the announcement t a becomes a more powerful<br />
instrument making a recourse to t less necessary.<br />
The difference between the statically optimal (i.e., evaluated at t $ ) profits<br />
of NBs and Bs<br />
g NB$ − g B$ = (1 + c x) 2<br />
− δ, (18)<br />
2c x (2π − 3) 2<br />
is likewise increasing in π. Modulo δ, the profit of the NBs is always higher<br />
than the profit of the Bs whenever π>0, reflecting the fact that the latter<br />
make a systematic error about the true value of t. The profit of the Bs,<br />
however, can exceed the one of the NBs if the prediction costs δ are high<br />
enough.<br />
Since t a$ is constant and t $ decreasing in π and since t $ ≤ t a$ , the<br />
difference t a$ − t $ , increases with π. Therefore, the difference Eq. (18)<br />
between the profits of the NBs and Bs is also increasing in the difference<br />
between the announced and the implemented tax, t a$ − t $ ,
110 Chirstophe Deissenberg and Pavel Ševčík<br />
Figure 1. The profits of the Bs (lower curve), of the NBs (middle curve), and the<br />
regulator’s utility (upper curve) as a function of π for δ = 0.00025, c x = 0.6, p = 2.<br />
As one would intuitively expect, the regulator’s instantaneous utility φ<br />
too, increases with π. Most remarkable, however, is the fact that the profits of<br />
both the Bs and NBs are also increasing in π (see Fig. 1 for an illustration).<br />
An intuitive explanation is the following: For π = 0 the outcome is the<br />
Nash solution of a game between the NBs and the regulator. This outcome<br />
is not efficient leaving room for Pareto-improvement. As π increases, the<br />
problem tends towards the solution a standard optimization situation with<br />
the regulator as the sole decision-maker. This solution is efficient in the sense<br />
that it does maximize the regulator’s objective function under the given<br />
constraints. Since the objectives of the regulator are not extremely different<br />
of the objectives of the firm, moving from the inefficient game-theoretic<br />
solution to the efficient control-theoretic solution one improves the position<br />
of all actors. Specifically, as π increases, the Bs enjoy the benefits of a lower<br />
t, while the investment remains fixed at t a$ . Likewise, the NBs benefit from<br />
the decrease in t. And R also benefits, because a raising number of Bs<br />
are led by R’s announcement to invest more than they would, otherwise<br />
and, subsequently, to produce more, to make higher profits, and to pollute<br />
subsequently, to produce more, to make higher profits, and to pollute less.<br />
In dynamic settings, however, the impact of the tax announcements<br />
on the π dynamics may incite the regulator to reduce the spread between<br />
t and t a , and in particular to choose t a
Cheap-talk Multiple Equilibria and Pareto 111<br />
a high proportion of Bs to a low one, since it has one more instrument<br />
(namely, t a ) to influence the Bs than to regulate the NBs. A high spread<br />
t a − t, however, implies that the profits of the Bs will be low compared<br />
to those of the NBs. This, by Eq. (6), reduces the value of ˙π and leads<br />
over time to a lower proportion of Bs, diminishing the instantaneous utility<br />
of R. Therefore, in the dynamic problem, R will have to find an optimal<br />
compromise between choosing a high value of t a , which allows a low value<br />
of t and insures the regulator a high instantaneous utility, and choosing a<br />
lower one, leading to a more favorable proportion of Bs in the future. We<br />
are going to explore these mechanisms in more details in the next section.<br />
4. The Dynamic <strong>Optimization</strong> Problem<br />
4.1. Formulation<br />
Being atomistic, the firms do not recognize that their individual decisions<br />
to choose the B or NB prediction technology influence π over time. Assume<br />
that they consider π as constant. The regulator’s optimization problem is<br />
then:<br />
max (t a ,t) 0 ≤ t a (τ), t(τ)<br />
(19)<br />
such that Eqs.(6), (7), (11), (13), and π 0 = π(0).<br />
Using<br />
g := p − c x(1 + 2πt a )<br />
1 + c x (3 − 2π) ,<br />
h := p − t<br />
(20)<br />
.<br />
2c x<br />
The Hamiltonian of the above problem is given by<br />
H(t a ,t,π,λ)= h + π<br />
[h −<br />
(p − t)2<br />
4c x<br />
+ t(h − t a ) − ta2<br />
2<br />
(p − t)2<br />
+ (1 − π)<br />
[h − + t(h − g) − 1 4c x<br />
2 g2 − δ<br />
+ t[π(h − t a ) + (1 − π)(h − g)]<br />
− [π(h − t a ) + (1 − π)(h − g)] 2<br />
+ λπ(1 − π)β<br />
[tt a − ta2<br />
2 − th + 1 ]<br />
2 h2 + δ ,<br />
where λ is the co-state.<br />
]<br />
]
112 Chirstophe Deissenberg and Pavel Ševčík<br />
The first order maximization condition with respect to t a and t yields<br />
for the dynamically optimal values*<br />
t a∗ = (p − c x) + [c x (3 − 2π) + 1](1 + c x ) 2<br />
, (21)<br />
(1 + 3c x )[βλ(1 − π) + K]<br />
with<br />
t ∗ = (p − c x) + 2c x (1 + c x )π[c x (βλ(1 + 3c x )(1 − π) + 1)]<br />
(1 + 3c x )[βπ(1 − π) + K]<br />
K = 1 − 6c 3 x β2 λ 2 (1 − π) 2 π + 2c x [2 + 2βλ(1 − π)π]<br />
+ cx 2 [3 − 2π + βλ(1 − π)(3 − 2βλ(1 − π)π)]. (22)<br />
According to Pontriagin’s maximum principle, an optimal solution<br />
(π(τ), (λ)) has to satisfy the state dynamics (6), evaluated at (t a∗ ,t ∗ ) plus:<br />
˙λ = ρλ − ∂H(ta∗ (π, λ), t ∗ (π, λ), π, λ)<br />
(23)<br />
∂π<br />
0 = lim<br />
τ→∞ e−ρτ λ(τ). (24)<br />
Due to the highly non-linear structure of the optimization problem an<br />
explicit derivation of the optimal (π(τ), λ(τ)) seems infeasible. Hence, we<br />
used Mathematica r○ to carry out a numerical analysis of the canonical<br />
system Eqs. (6)–(23) in order to shed light on the optimal solution.<br />
4.2. Numerical Results<br />
Depending upon the parameter values, the model can have either two stable<br />
equilibria separated by a so-called Skiba point, or a unique stable equilibrium.We<br />
discuss first the more interesting case of multiple equilibria, before<br />
carrying a summary of sensitivity analysis and briefly addressing the question<br />
of the bifurcations, leading to a situation with a situation with unique<br />
equilibrium. Unless and otherwise specified, the results presented pertain<br />
to the reference parameter configuration c x = 0.6, p = 2, δ = 0.00025 that<br />
underlies Fig. 1, together with ρ = 1, 0.0015. These are robust, with respect<br />
to sufficiently small parameter variations: The same qualitative insights<br />
would be obtained for a compact set of alternative parameter configurations.<br />
From the onset, let us stress the fundamental mechanism that underlies<br />
most of the results. Ceteris paribus, R would like to face as many Bs as<br />
possible, since dφ/dπ >0. If the prediction costs δ are sufficiently high,<br />
the profit difference g NB$ − g B$ see Eq. (18), will be negative at the current
Cheap-talk Multiple Equilibria and Pareto 113<br />
value of π. In this case, by Eq. (6), implementing the statically optimal<br />
actions t a$ , t $ suffices (albeit not optimally so) to insure that π increases<br />
locally. In fact, for δ sufficiently large, the thus controlled system will<br />
globally converge to a situation where there are only believers, π = 1.<br />
However, if δ is small, the regulator cannot implement the actions t a$ , t $<br />
since in that case g NB
114 Chirstophe Deissenberg and Pavel Ševčík<br />
Figure 2. Phase diagram of the canonical system, c x = 0.6, p = 2, δ = 0.00025,<br />
β = 1, and ρ = 0.0015.<br />
E U<br />
Table 1. Profits, welfare, and taxes at E U<br />
and E L , c x = 0.6, p = 2, δ = 0.0025, β = 1,<br />
p = 0.0015.<br />
g φ t a<br />
t<br />
E U 1 2.25 1.07 0.12<br />
E L 0.98 2.18 N.A. 0.62<br />
The Pareto-superiority of E U is illustrated in Table 1 (remember that at<br />
the equilibrium g N = g NB ).<br />
The situation captured in Fig. 2 (an unstable equilibrium surrounded<br />
by two stable ones) implies the existence of a threshold such that it is<br />
optimal for R to follow a policy leading in the long-run towards the stable<br />
equilibrium E L = (0,λ L ), whenever the initial value π 0 of π is inferior to<br />
the threshold value, π 0
Cheap-talk Multiple Equilibria and Pareto 115<br />
Whenever π 0 π s lies in the following<br />
fact: The “deviation costs” that the regulator will have to accept initially in<br />
order to steer the system towards the more favorable E U are so high that the<br />
net benefit of going to E U would be less than the one obtained by letting<br />
the system optimally converge towards the inferior E L .<br />
The above results suggest that the policy of an environmental regulator<br />
maybe very different depending on the fraction of the population that trusts<br />
it when it initiates a new policy. If the initial confidence is high, the costs of<br />
building a large fraction of Bs are compensated in the long-run by accrued<br />
benefits. However, if the regulator’s trustworthiness is low from the onset,<br />
its interest is to exploit its initial credibility for making short-term gains,<br />
although this ultimately leads to an inferior situation where nobody trusts<br />
the regulator.<br />
As already mentioned, the dynamics of the canonical system Eqs. (6)–<br />
(23) at the unstable middle equilibrium E M form a spiral. Therefore, the<br />
threshold value π s is typically close to, but distinct from π M (i.e., π M ̸= π s )<br />
and the threshold takes the particularly challenging form of a Skiba point.<br />
No local analytical expression exists that characterizes a Skiba point. Thus,<br />
Skiba points must be computed numerically. This was not done in this paper.<br />
For a reference article on thresholds and Skiba points in economic models,<br />
see Deissenberg et al. (2004).<br />
5. Sensitivity Analysis and Bifurcations Towards<br />
a Unique Equilibrium<br />
Numerical analyses show that an increase in the firms’ flexibility (i.e., an<br />
increase in β) shifts the stable equilibrium E U to the right and the unstable
116 Chirstophe Deissenberg and Pavel Ševčík<br />
equilibrium E M as well as the Skiba point π s to the left. In other words,<br />
if the firms react quickly to profit differences, the regulator will follow a<br />
policy that leads to the upper equilibrium E U even if the initial proportion<br />
of Bs is small. Indeed, the high speed of reaction of the firms means that the<br />
accumulated costs incurred by the regulator on the way to E U will be small<br />
and easily compensated by the gains around and at E U . Moreover, R does<br />
not have to make Bs much better off than NBs to insure a fast reaction. As a<br />
result, for β large, the equilibrium E U is characterized by a large proportion<br />
of Bs and thus insures high stationary gains to all actors.<br />
Not surprisingly, an increase of the discount factor ρ has the opposite<br />
effect. Impatient regulators want to build up confidence, only if the initial<br />
proportion of Bs is high. The resulting equilibrium value π U will be relatively<br />
low, since the time and efforts needed now for an additional increase<br />
in the stock of Bs weighs heavily compared to the expected future benefits.<br />
An increase of the prediction cost δ finally, shifts E M to the left and<br />
E U to the right: It is easier for R to incite the firms to use the B strategy,<br />
favoring the convergence to equilibrium with a large number of believers.<br />
The scenario with two stable equilibria separated by a Skiba point can<br />
change qualitatively, when the parameter values are sufficiently altered.<br />
For small values of β and/or δ and/or for large values of ρ, the equilibrium<br />
E U collides with E M and disappears. Only one (stable) equilibrium<br />
remains, E L . Thus, in the long-run, the proportion of believers tends to zero.<br />
On the other hand, for large values of δ trying to outguess the regulator is<br />
never optimal. The only remaining equilibrium is the one where everybody<br />
is a believer.<br />
6. Conclusion<br />
The starting point of this paper is a static situation, where standard rational<br />
behavior by all agents leads to a Pareto-inferior outcome, although<br />
there is no fundamental conflict of interest either among the different firms<br />
or between the firms and the regulator, and although the firms are perfectly<br />
informed and perfectly predict the regulator’s actions. In addition<br />
to the environmental problem studied here, such a situation is common for<br />
instance in monetary economics, in disaster relief, patent protection, capital<br />
levies, or default on debt.<br />
The present paper and previous work strongly suggest that, in such a<br />
situation, the existence of believers who take the announcements of a macroeconomic<br />
regulator at face value typically, Pareto-improve the economic
Cheap-talk Multiple Equilibria and Pareto 117<br />
outcome. This property hinges crucially on the fact that the private sector<br />
is atomistic, and thus, that the single agents do not anticipate the collative<br />
impact of their individual decisions. Moreover, it requires a relatively<br />
benevolent regulator, whose objective function does not differ too drastically<br />
from the one of the private agents.<br />
To introduce dynamics, we assumed that the proportion of believers<br />
and non-believers in the economy changes over time as a function of the<br />
difference in profits for the two types of firms. The change obeys a word-ofmouth<br />
process driven by the relative success of the one or the other private<br />
strategy, to believe or to predict. It is supposed that the regulator recognizes<br />
its ability to influence the word-of-month dynamics, by an appropriate<br />
choice of actions and is interested not only in its instantaneous but also<br />
in its future losses. It is shown that, when the firms react fast enough to the<br />
difference in profits of Bs and NBs, a sufficiently patient regulator may use<br />
incorrect announcements to Pareto-improve the economic outcome compared<br />
to the situation where all firms are non-believers that perfectly predict<br />
the regulator’s actions. A particularly interesting case arise, when a stable<br />
Pareto-superior equilibrium with a positive fraction of believers co-exists<br />
with the (likewise stable) perfect prediction but inferior equilibrium where<br />
all firms are non-believers.<br />
In this case, the optimal policy of the regulator (to converge towards<br />
the one or the other equilibrium) depends upon the initial proportion of<br />
believers in the population. This history dependence of the solution suggests<br />
a promising meta-analysis of the importance of good reputation based on the<br />
previous good governance, in situations where the same macro-economic<br />
policy-maker is confronted with recurrent, distinct decision-making issues.<br />
These results of this paper are obtained under the assumption that the<br />
non-believers do not exploit the fact that the regulator announced tax differs<br />
from the equilibrium announcement for the static game. Likewise, in the<br />
model, the non-believers do not attempt to predict the dynamics of the number<br />
of believers.We leave to future research, the non-trivial task of exploring<br />
the consequences of a relaxation, of these two restrictive hypotheses.<br />
References<br />
Abrego, L and C Perroni (1999). Investment subsidies and time-consistent environmental<br />
policy. Discussion Paper, Coventry: University of Warwick, 1–35.<br />
Barro, R and D Gordon (1983). Rules, discretion and reputation in a model of monetary<br />
policy. Journal of Monetary <strong>Economic</strong>s 12, 101–122.
118 Chirstophe Deissenberg and Pavel Ševčík<br />
Batabyal, A (1996a). Consistency and optimality in a dynamic game of pollution control I:<br />
competition. Environmental and Resource <strong>Economic</strong>s 8, 205–220.<br />
Batabyal, A (1996b). Consistency and optimality in a dynamic game of pollution control II:<br />
monopoly. Environmental and Resource <strong>Economic</strong>s 8, 315–330.<br />
Dawid, H (1999). On the dynamics of word of mouth learning with and without anticipations.<br />
Annals of Operations Research 89, 273–295.<br />
Dawid, H and C Deissenberg (2005). On the efficiency-effects of private (dis)-trust in the<br />
government. Journal of <strong>Economic</strong> Behavior and Organization 57, 530–550.<br />
Dawid, H, C Deissenberg and P Ševčík (2005). Cheap talk, gullibility and welfare in an<br />
environmental taxation game. In: Dynamic Games: Theory and Applications, A Haurie<br />
and G Zaccour (eds.), Amsterdam: Kluwer, 175–192.<br />
Deissenberg, C and F Alvarez Gonzalez (2002). Pareto-improving cheating in a monetary<br />
policy game. Journal of <strong>Economic</strong> Dynamics and Control 26, 1457–1479.<br />
Deissenberg, C, G Feichtinger, W Semmler and F Wirl (2004). Multiple equilibria, history<br />
dependence, and global dynamics in intertemporal optimization models. In <strong>Economic</strong><br />
Complexity: Non-Linear Dynamics, Multi-agents Economies, and Learning,<br />
W Barnett, C Deissenberg and G Feichtinger (eds.), ISETE, Vol. 14, Amsterdam:<br />
Elsevier, 91–122.<br />
Dijkstra, B (2002). Time consistency and investment incentives in environmental policy.<br />
Discussion Paper 02/12.U.K.: School of <strong>Economic</strong>s, University of Nottingham, 2–12.<br />
Gersbach, H and A Glazer (1999). Markets and regulatory hold-up problems. Journal of<br />
Environmental <strong>Economic</strong>s and Management 37, 151–164.<br />
Hofbauer, J and K Sigmund (1998). Evolutionary Games and Population Dynamics.<br />
Cambridge: Cambridge University Press.<br />
Kydland, F and E Prescott (1977). Rules rather than discretion: the inconsistency of optimal<br />
plans. Journal of Political Economy 85, 473–491.<br />
Marsiliani, L and T Renstrom (2000). Time inconsistency in environmental policy: tax<br />
earmarking as a commitment solution. <strong>Economic</strong> Journal 110, C123–C138.<br />
McCallum, B (1997). Critical issues concerning central bank independence. Journal of<br />
Monetary <strong>Economic</strong>s 39, 99–112.<br />
Petrakis, E and A Xepapadeas (2003). Location decisions of a polluting firm and the time<br />
consistency of environmental policy. Resource and Energy <strong>Economic</strong>s 25, 197–214.<br />
Simaan, M and JB Cruz (1973a). On the Stackelberg strategy in nonzero-sum games. Journal<br />
of <strong>Optimization</strong> Theory and Applications 11, 533–555.<br />
Simaan, M and JB Cruz (1973b). Additional aspects of the Stackelberg strategy in nonzerosum<br />
games. Journal of <strong>Optimization</strong> Theory and Applications 11, 613–626.<br />
Vallee, T (1998). Comparison of different Stackelberg solutions in a deterministic<br />
dynamic pollution control problem. Discussion Paper, LEN-C3E. France: University of<br />
Nantes, 1–30.<br />
Vallee, T, C Deissenberg and T Basar (1999). Optimal open loop cheating in dynamic<br />
reversed linear-quadratic Stackelberg games. Annals of Operations Research 88,<br />
247–266.
Chapter 4<br />
Modeling Business Organization
This page intentionally left blank
Topic 1<br />
Enterprise Modeling and Integration:<br />
Review and New Directions<br />
Nikitas Spiros Koutsoukis<br />
Hellenic Open University, and Democritus University<br />
of Thrace, Greece<br />
1. Introduction<br />
Enterprise modeling (EM), Enterprise integration (EI), and Enterprise<br />
integration modeling (EIM) are concerned with the definition, analysis,<br />
re-design and integration of information, human capital, business process,<br />
processing of data and knowledge, software applications and information<br />
systems regarding the enterprise and its external environment to achieve<br />
advancements in organization performance (Soon et al., 1997).<br />
Enterprise modeling and Enterprise integration still are still ongoing<br />
fields of research where the background and interests of contributors in<br />
these areas are heterogeneous, including disciplines like mechanical manufacturing,<br />
business process re-engineering, and software engineering.<br />
However, the following two definitions (Vernadat, 1996) are representative<br />
for current practices:<br />
• Enterprise modeling is concerned with the explicit representation of<br />
knowledge regarding the enterprise in terms models.<br />
• Enterprise integration consists of breaking down organizational barriers<br />
to improve synergy within the enterprise so that business goals are<br />
achieved in a more productive way.<br />
The main objective of this paper is to review the literature on EM and integration<br />
and to provide a concise description of related topics. The paper also<br />
discusses EM and integration challenges, and rationale as well as current<br />
tools and methodologies for deeper analysis.<br />
121
122 Nikitas Spiros Koutsoukis<br />
2. Model and Enterprise Terminology<br />
Model and enterprise are the two key terms that appear throughout this<br />
paper. An enterprise can be perceived as “a set of concurrent processes executed<br />
on the enterprise means (resources or functional entities) according<br />
to the enterprise objectives and subject to business or external constraints”<br />
(Vernadat, 1996). A model can be defined as “a structure that a system<br />
can use to simulate or anticipate the behaviour of something else” (Hoyte,<br />
1992). This is a generalization of a definition given by Minsky (1968):<br />
“a model is a useful representation of some subject. It is a (more or less<br />
formal) abstraction of a reality (or universe of discourse) expressed in terms<br />
of some formalism (or language) defined by modeling constructs for the<br />
purpose of the user”.<br />
Instead of attempting to form a definition, which may be redundant, we<br />
will mention some important features of enterprise models instead.<br />
• An enterprise model provides a computational representation of the<br />
structure, activities, processes, information, resources, people, behavior,<br />
goals, and constraints of a business, government, or other enterprise.<br />
It can be both descriptive and definitional-spanning what is and what<br />
should be. The role of an enterprise model is to achieve model-driven<br />
enterprise design, analysis, and operation (Fox, 1993).<br />
• Enterprise models must either be represented in the mind of human beings<br />
or in computers (Christensen et al., 1995).<br />
• All enterprise models are built with a particular purpose in mind. Depending<br />
on the purpose, the model will emphasize different types of information<br />
at different levels of detail.<br />
3. Enterprise Modeling<br />
During the last decades, EM has been partly addressed by several<br />
approaches to the same problem of making the enterprise more efficient.All<br />
these approaches have included the construction of symbolic representations<br />
of aspects of the real world, using various notations and techniques that<br />
commonly could be denoted “conceptual modeling” (Solvberg and Kung,<br />
1993). The idea of EM, as an extension of the standard for the exchange<br />
of product model data (STEP) product modeling approach, contributes to<br />
integrating design (which focuses on product information) and construction<br />
(focusing on activity and resource information) (Kemmerer, 1999).
Enterprise Modeling and Integration 123<br />
In the 1980s and during the early 1990s, when database technology was<br />
applied on a larger scale, the need for better application design and development<br />
methods became evident (Martin, 1982; Sager, 1988). Different<br />
modeling methodologies, like the structured analysis and design technique<br />
(SADT), were developed. A narrow focus on a single department or task<br />
during development often caused operational problems across application<br />
boundaries. The need for making more complete models aroused, integrating<br />
operations across departmental or functional boundaries.<br />
The problem in EM regarding the manufacturing of EI and control was<br />
to ensure timely execution of business processes on the functional entities<br />
of the enterprise (i.e., human and technical agents) to process enterprise<br />
objects. Processes are made of functional operations and functional operations<br />
are executed by functional entities. Objects flowing through the enterprise<br />
can be information entities (data) as well as material objects (parts,<br />
products, tools, etc.) (Rumbaugh, 1993).<br />
In order to design the flexible information and communication<br />
infrastructure needed, different enterprise models are useful, and the model<br />
may in fact become a part of the integrated enterprise itself. The trend<br />
now is away from the first single perspective enterprise models, which was<br />
only focusing on the product data information handling, to more complete<br />
enterprise models covering both informational and human aspects of the<br />
organization (Fox, 1993).<br />
During the 1990s, the question aroused on how to integrate different<br />
departments in an enterprise, and how to connect the enterprise with its<br />
external environment i.e., suppliers and customers, to improve co-operation<br />
and logistics. During the 1990s, conglomerate industries took a more developmental<br />
approach, and the research area of EI emerged. Enterprise modeling<br />
is clearly a pre-requisite for EI as all things to be integrated and<br />
coordinated need to modeled to some extent (Petrie, 1992).<br />
Since 1990, major R&D programs for computer-integrated manufacturing<br />
(CIM) have resulted in the following tools/methods or architectures<br />
for EM:<br />
(1) IDEF modeling tools: These are complementary modeling tools developed<br />
by the integrated computer aided manufacturing (ICAM) project,<br />
introducing IDEF0 for functional modeling, IDEF1 for information<br />
modeling, IDEF2 for simulation modeling, IDEF3 for business process<br />
modeling, IDEF4 for object modeling, and IDEF5 for ontology<br />
modeling. IDEF models are mainly used for requirements definitions.<br />
(2) Generic Artificial Intelligence (GRAI) integrated methodology: It is a<br />
methodology which uses the GRAI grid, GRAI nets, developed by the
124 Nikitas Spiros Koutsoukis<br />
GRAI laboratories of France and includes modeling IDEF0, and group<br />
technology to model the decision making processes of the enterprise.<br />
The combination is called GRAI integrated methodology (GIM).<br />
(3) Computer integrated manufacturing open system architecture<br />
(CIM-OSA): CIMOSA has been developed to assist enterprises adapt to<br />
internal and external environmental changes so as to face global competition<br />
and improve themselves regarding product prices, product quality,<br />
and product delivery time. It provides four different types of views<br />
of the enterprise: modeling function view, information view, resource<br />
view, and organization view (Beeckman 1993; Vernadat, 1996).<br />
(4) Purdue enterprise reference architecture (PERA): It is a methodology<br />
developed at Purdue University in 1989 which focuses on separating<br />
the human-based functions of an enterprise from the manufacturing<br />
and information functions (Williams, 1994).<br />
(5) Generalized enterprise reference architecture and methodology<br />
(GERAM): It is a generalization of CIMOSA, GIM, and PERA. The<br />
general concepts, identified and defined in this reference methodology,<br />
consist of methodological guidelines for enterprise engineering (from<br />
PERA and GIM), life-cycle guidelines (from PERA), and model views<br />
modeling (e.g., CIMOSA constructs).<br />
Modeling method relies upon a specific purpose defining its finality, i.e.,<br />
the goal of the modeler. This finality has a direct influence on the definition<br />
of modeling method. We adopt the position that any enterprise is made of a<br />
large collection of concurrent business processes, executed by an open set<br />
of functional entities (or agents) to achieve business objectives (as set by<br />
management). Enterprise modeling and integration is essentially a matter<br />
of modeling and integrating these process and agents (Kosanke and Nell,<br />
1997; Ladet and Vernadat, 1995).<br />
The prime goal of an EM approach is to support analysis of an enterprise<br />
rather than to model the entire enterprise, even though this is theoretically<br />
possible. In addition, another goal is to model relevant business process and<br />
enterprise objects concerned with business integration.<br />
According to Vernadat (1996), the aim of EM is to provide:<br />
• a better perspective of the enterprise structure and operations;<br />
• reference methods for enterprise engineering of existing or new parts of<br />
the enterprise both in terms of analysis, simulation, and decision-making<br />
and<br />
• a model in order to manage efficiently the enterprise operations.
Whereas, the main motivations for EM are:<br />
Enterprise Modeling and Integration 125<br />
• better understanding of the structure and functions of the enterprise;<br />
• managing complexity of operations;<br />
• creating enterprise knowledge and know-how for all;<br />
• efficient management of processes;<br />
• enterprise engineering and<br />
• EI itself.<br />
4. Enterprise Integration<br />
Enterprise integration was discussed since the early days of computers<br />
in industry and especially in the manufacturing industry with CIM as the<br />
acronym for operations integration (Vernadat, 1996). In spite of the different<br />
understandings of the scope of integration in CIM, it was always understood<br />
for information integration across or at least parts of the enterprise. Information<br />
integration essentially consists of providing the right information,<br />
to the right people, at the right place, at the right time (Norrie et al., 1995).<br />
Information that is available to the right people, at the right place, at the<br />
right time, and that enables them to act quickly and decisively, is a crucial<br />
requirement of enterprises that intend to grow and remain profitable. The<br />
intelligent delivery of information must obtain maturity levels consistent<br />
with enterprise goals and must employ the appropriate technology to facilitate<br />
this purpose. In today’s enterprise environment, users must consolidate<br />
data and deliver it as useful information and knowledge that empowers the<br />
right business decisions.<br />
With this understanding the different needs in EI can be identified:<br />
(1) Identify the right information to gain knowledge: Elicit knowledge from<br />
gathering information from the intra-enterprise activities and use models<br />
to represent this knowledge regarding product attributes and information,<br />
process management issues, and the decision-making process.<br />
(2) Provide the right information at the right place:The need for implementing<br />
information sharing systems and integration platforms to support<br />
information transaction across heterogeneous environments regarding<br />
hardware, different operating systems, and the legacy systems. This<br />
is a step towards inter-enterprise integration overcoming barriers and<br />
provides connection between enterprise environments by linking their<br />
operations to create extended and virtual enterprises.
126 Nikitas Spiros Koutsoukis<br />
(3) Up-date the information in real time to reflect the actual state of the<br />
enterprise operation: enterprises must be able to adopt changes regarding<br />
both their external environment i.e., technology, legislation, and<br />
their internal environment i.e., production operations in order to protect<br />
the enterprise goal and objective from becoming obsolete.<br />
(4) Co-ordinate business processes: This requires decisional capabilities<br />
and know-how within the enterprise for real-time decision support<br />
and evaluation of operational alternatives based on business process<br />
modeling.<br />
The aims of EI are:<br />
• to enable communication among the various functional entities of the<br />
enterprise;<br />
• to provide interoperability of IT applications (plug-and-play approach)<br />
and<br />
• to facilitate coordination of functional entities for executing business<br />
processes so that they synergistically fulfill enterprise goals.<br />
The main motivation for EI is to make possible true interoperability<br />
among heterogeneous, remotely located, systems within, or outside the<br />
enterprise in an independent IT vendor environment.<br />
Enterprise integration, especially for networked companies and large<br />
manufacturing enterprises is becoming a reality. The issue is that EI is both<br />
an organizational and a technological matter. The organizational matter<br />
is partly understood so far whereas the technological matter lies with the<br />
major advances over the last decade. It concerns mostly computer communications<br />
technology, data exchange formats, distributed databases, object<br />
technology, Internet, object request brokers (ORB such as OMG/CORBA),<br />
distributed computing environments (such as OSF/DCE and MS DCOM,<br />
and now Java to enterprise edition and execution environments (J2EE).<br />
Some important projects developing integrating infrastructure (IIS)<br />
technology for manufacturing environments include:<br />
• CIMOSA IIS: The CIMOSA integrated infrastructure (IIS) enables<br />
business process control in a heterogeneous IT environment. In IIS, an<br />
enterprise is composed of functional entities, which are connected by a<br />
communication system. Functional entities can be a machine, a software<br />
application, or a human (Vernadat, 2002).<br />
• AIT IP: This is an integration platform developed as an AIT project and<br />
based on the CIMOSA IIS concepts (Waite, 1998).
Enterprise Modeling and Integration 127<br />
• OPAL: This is also an AIT project that has proved that EI can be achieved<br />
in design and manufacturing environments using existing ICT solutions<br />
(Bueno, 1998).<br />
• NIIIP (National industrial information infrastructure protocols) (Bolton<br />
et al., 1997).<br />
However, EI suffers from inherent complexity of the field, lack of established<br />
standards, which sometimes happen after the fact, and the rapid and<br />
unstable growth of the basic technology with a lack of commonly supported<br />
global strategy. The EI modeling research community is growing at<br />
a very rapid pace (Fox and Gruninger, 2003). The EIM research so far has<br />
been mostly performed in large research organizations. A few pilot studies<br />
have also been done in large organizations. Hunt (1991) suggests that EIM<br />
is most beneficial to large industrial and government organizations. The<br />
emphasis of current EIM research on large organizations is risky, and may<br />
even have hampered the growth of the field. Enterprise wide modeling is<br />
difficult in large organizations for two reasons: there is seldom any consensus<br />
on exactly how the organization functions and the modeling task may<br />
be so big that by the time any realistic model is built, the underlying domain<br />
may have changed. Whatever these terms mean, we have to seek solutions<br />
to solve industry problems, and this applies to EM and integration.<br />
5. A Framework Setting for EI Modeling<br />
Enterprise integration is a multifaceted task employed for addressing challenges<br />
stemming from various aspects of the globalized economy. The contemporary<br />
business world, economically, politically, and technologically is<br />
a fast-changing world and most if not all enterprises, from small to large,<br />
are continually challenged to evolve as rapidly as the world around them<br />
evolves. Enterprise integration allows an enterprise to function coherently<br />
as a whole and to shift its primary value chain activities in tandem with<br />
its secondary value chain activities according to the challenges or requirements<br />
posed by the enterprise environment or goals set by the enterprise<br />
itself, and most frequently a combination of both (Giachetti, 2004). In this<br />
context, the value chain’s primary and secondary activities are used simply<br />
as a well-known functional-economic view of the enterprise in order to<br />
communicate coherently, as opposed to a semantically accurate or complete<br />
view of the enterprise (Porter, 1985).
128 Nikitas Spiros Koutsoukis<br />
The multiple facets of EI refer to the multiple dimensions, entities, and<br />
constituents where integration is primarily aimed at, and through which<br />
enterprise integration is achieved.<br />
Enterprise integration dimensions represent infrastructural views, upon<br />
which integration is founded, built and subsequently achieved. Some of the<br />
most commonly used integration dimensions are the following:<br />
• the data and information dimension;<br />
• the processes dimension;<br />
• the economic activity or business dimension and<br />
• the technology or technical dimension.<br />
Because these dimensions can easily be extended throughout the enterprise<br />
regardless of the enterprise functions, they are frequently used as<br />
fundamental reference points for EI efforts (Lankhorst, 2005).<br />
Each of these dimensions has been used successfully for seeking out and<br />
achieving EI. For instance, in the data and information dimension, which<br />
is mainly the realm of corporate information systems, data warehousing<br />
and enterprise resource planning (ERP), have been used successfully to<br />
introduce enterprise-wide, unified views of data, information and related<br />
analyses. In the processes dimension, business process management (BPM)<br />
or re-engineering (BPR) have been successfully applied in developing and<br />
refining old, and new processes which are more integrated and efficient<br />
across the enterprise. In the economic or business dimension, business<br />
strategy models, like the core-competency theory, balanced-score cards,<br />
or economic value-added models like the activity-based costing (ABC)<br />
are also used successfully to integrate across the enterprise (Kaplan and<br />
Norton, 1992; Prahalad and Hamel, 1990). In the technology or technical<br />
dimension, the key requirement is for interoperability of systems, hardware,<br />
and software. The success with which many multinational enterprises<br />
operate shows that certain milestones have been reached in this<br />
direction.<br />
Enterprise entities refer to function-oriented enterprise constructs<br />
that are typically goal-oriented, collectively contributing to achieving<br />
the enterprise mission. Entities form hierarchical views of the enterprise,<br />
and frequently tend to match formal or informal structures as in<br />
most organizational charts. From a functional perspective, entities refer<br />
to key organizational activities or divisions, such as administration, production,<br />
accounting, sales, marketing, procurement, and R&D. As is well<br />
known, entities can also be organized in terms of authority or delegation
Enterprise Modeling and Integration 129<br />
layers, communication layers, goal-contribution layers, geographically<br />
based activities and so forth.<br />
The sole purpose for having such functional-oriented constructs is to<br />
decompose the enterprise into smaller, more specific, and more manageable<br />
parts. The contribution of each entity to the enterprise objectives is thus<br />
readily identified and narrowed down in scope, horizontally or vertically.<br />
It is only through the use of one or more dimensions that the entities are<br />
re-combined in an enterprise-wide integration model.<br />
Enterprise constituents or constituent groups refer to the human factor<br />
which is ultimately responsible for setting out, applying and achieving EI.<br />
The most prevalent constituent groups are the following:<br />
• Stakeholders<br />
• Decision-makers<br />
• Employees<br />
which are internal to the enterprise, and<br />
• Suppliers and<br />
• Clients<br />
which are external to the enterprise.<br />
Some of the main pitfalls in EI lie at the constituents’ level. This is<br />
because the constituents are, in one way or another, autonomous or semiautonomous<br />
agents acting primarily in their own interest, which is defined<br />
mainly by their position in enterprise coordinates terms. We consider the<br />
following fundamental dimensions as the main enterprise coordinates for<br />
positioning the constituents:<br />
• level of authority and<br />
• immediacy of contribution to goals (from strategic to operational, abstract<br />
to specific)<br />
which can also be inverted to<br />
• level of responsibility and<br />
• level of activity (from strategic to operational, or abstract to specific)<br />
The higher the level of authority, the more strategic or abstract are the<br />
actions that the constituents take in enterprise terms. At this level, constituents<br />
have a bird’s-eye view of the enterprise and find it easier to match<br />
organizational outcomes to personal interest i.e., the bird’s-eye view makes<br />
it easier to view the enterprise as a single entity. On the other hand, the
130 Nikitas Spiros Koutsoukis<br />
lower the level of authority the more operational or specific the actions<br />
that the constituents take in enterprise terms. At this level, constituents find<br />
it harder to consider a single-entity view of the organization, and hence<br />
organizational outcomes are less easily matched to personal interests.<br />
It follows that inconsistencies in the constituents’ perspectives have the<br />
ability to cancel out some of the main benefits that the other two facets bring<br />
forward, namely the unified view of the enterprise as single entity and the<br />
immediacy with which operational activities contribute to strategic goals.<br />
Inconsistencies in constituents’ perspectives are being addressed mainly<br />
through leadership and human resource practices, which aim to put in place<br />
an organizational culture, or otherwise a unified perception of what the<br />
enterprise stands for, its values and its practices among other things. Mission<br />
statements, codes of practice, and team building exercises are some of the<br />
most frequently used tools for scoping and setting out a unified, constituent<br />
view of the enterprise.<br />
Thus, a strong requirement is surfacing for developing new EI modeling<br />
frameworks, which are able to incorporate heterogeneous modeling constructs,<br />
such as the dimensions, entities or constituents considered above.<br />
We find that some of the key requirements are set out as follows:<br />
(a) Interchangeable dimensions: EI models should be able to shift between<br />
the dimensions considered above, without loss of context.<br />
(b) Use of hybrid modeling constructs: EI models should be able to utilize<br />
interchangeably modeling constructs for representing any type of enterprise<br />
activity unit, including procedural units, physical or logical units,<br />
inputs or outputs, tangible and intangible processes, decision-making<br />
points, etc.<br />
(c) Ability to consolidate or analyze to successive levels of detail: this is<br />
particularly useful for communicating enterprise structures throughout<br />
multiple echelons and hierarchy levels. For instance, stakeholders will<br />
most likely need access to a consolidated view of the integrated enterprise<br />
whereas employees are most likely to require a position-centred<br />
view of the enterprise, with detailed surroundings that are consolidated<br />
further away from the employee’s position.<br />
(d) Ability to interchange between declarative and goal-seeking states of<br />
the integration models: when in declarative mode, models work simply<br />
as formal structures, or simulators of how the integrated enterprise<br />
operates. In goal-seeking mode, the model emphasizes improvements<br />
that can lead to goal attainment or “satisfying,” to use Simon’s term<br />
(Simon, 1957).
Enterprise Modeling and Integration 131<br />
Of course, these requirements can be analyzed further and in detail, but<br />
such a discussion would be beyond the scope of this paper. Still, we also<br />
find that these requirements are suffice to set out an agenda for further work<br />
in the domain of EI and EI modeling.<br />
6. Discussion<br />
At the current state of technology, we find that EM and integration are<br />
becoming familiar within the large enterprises that seem to understand the<br />
need of acquiring their concept. However, in accordance with Williams<br />
(1994) we also find that the reason for the moderate success and rather not<br />
so extended use of EM and integration initiatives include:<br />
(a) Cost: It is commonly accepted that such efforts are very expensive and<br />
it is rather difficult to argue on the benefits of the outcome from the<br />
beginning of each project.<br />
(b) Project size and duration: Indeed, such projects cannot be completed<br />
but they require time. Quite a large number of people will be involved<br />
and high levels of commitment on behalf of the enterprise are required.<br />
(c) Complexity: EI is a never-ending process and parameters such as the<br />
legacy systems within large enterprises resisting integration have to<br />
be seriously considered. There is the need for a global vision with a<br />
modular implementation approach.<br />
(d) Management support: Management facing high cost may resent to support<br />
such projects. It is very important that management be part of the<br />
project and be fully committed to it.<br />
(e) Skilled people: The human factor concerning high-skilled employees<br />
involved in the projects is vital for success. Unfortunately, due to limited<br />
training schemes experienced users are scarce.<br />
Nevertheless, the enthusiasm of the research community and the industry<br />
remains intact trying to overcome problems and difficulties by creating<br />
tools and adopting methodology that will make a difference. Nowadays,<br />
EM and integration mostly rely upon advanced computer networks, the<br />
worldwide web and Internet, integration platforms as well as ERPs, and<br />
data exchange formats (Chen and Vernadat, 2002; Martin et al., 2004).<br />
A big issue is that both EM and EI are nearly totally ignored by SME’s<br />
and there is still a long way to go before SME’s will master these techniques<br />
on their own and in this case, they have to quickly catch on the technology<br />
and define which tools they have to use.
132 Nikitas Spiros Koutsoukis<br />
Some researchers like Vernadat, Kosanke, Tolone, and Zeigler consider<br />
that this technology would better penetrate and serve any kind of<br />
enterprises if:<br />
• They identify users and a specific EIM problem. An example of an EIM<br />
problem is how to produce high-quality products at low cost.<br />
• They make a prototype EIM that addresses the problem, contributing in<br />
the decision-making process based on a simple initial model accepted by<br />
the majority of the users (Tolone et al., 2002).<br />
• There was a standard vision on what EM really depend and there was an<br />
international consensus on the underlying concept for the benefit of the<br />
business users (Kosanke et al., 2000).<br />
• There was a standard, user-oriented, interface in the form of a unified<br />
enterprise modeling language (UEML) based on the previous consensus<br />
to be available on all commercial modeling tools.<br />
• There were real EM and simulation tools commercially available in the<br />
market place taking into account function, information, resource, organization,<br />
and financial aspects of an enterprise including human aspects,<br />
exception handling, and process coordination. Simulation tools need to<br />
be configurable, distributed, and agent-based simulation tools (Vernadat<br />
and Zeigler, 2000).<br />
However, it is rather unlikely that a single EIM that will be equally<br />
applicable to all organizations. Each organization will have to develop its<br />
own strategy and philosophy (Fox and Huang, 2005). Despite the different<br />
approaches (frameworks) of EIM, we always have to bear in mind the<br />
following functional requirements of an EIM framework (Patankar and<br />
Adiga, 1995):<br />
(1) Decision orientation. An EI model based on any particular concept of<br />
an EM framework should support the decision-making process considering<br />
thoroughly all factors of the external and internal environment<br />
affecting the enterprise.<br />
(2) Completeness. EM frameworks should be able to represent all aspects<br />
of the enterprise such as business units, business data, information<br />
systems, and external partners i.e., suppliers, sub-contractors, and<br />
customers.<br />
(3) Appropriateness. EIM frameworks should be representing the enterprise<br />
at appropriate levels by providing models that can be understood<br />
by people with heterogeneous backgrounds.
Enterprise Modeling and Integration 133<br />
(4) Permanence. The EI model should remain consistent with the enterprise’s<br />
overall objectives even when the enterprise is in the process of<br />
changing due to management changes or other. <strong>Models</strong> should remain<br />
true and valid within the enterprise.<br />
Finally, the EIM frameworks should be simply structured, flexible,<br />
modular, and generic to ensure that the framework is applicable across different<br />
manufacturing operations and enterprises. In addition, the framework<br />
should relate to both manual and automatic processes. This will provide the<br />
basis for proper and complete integration.<br />
Modeling and integration always mean effort in time and costs.<br />
Nevertheless, to handle complexity in business environments we see the<br />
above-mentioned elements moving from “luxury articles” to “everyday<br />
necessities”.<br />
These are some of the challenges to be solved in future to build extended<br />
interoperable manufacturing enterprises as there is unlikely to be a single<br />
EIM methodology, which will be equally applicable to all organizations<br />
(Li and Williams, 2004). Each organization will have to develop its own<br />
strategy and philosophy. However, EIM promises to be a revolutionary<br />
technology.<br />
Finally, the success of EI efforts lies in the ability to iron out incoherencies<br />
and lack coordination between the multiple facets considered above.<br />
This is a complex and challenging task, and lies at the heart for EI modeling.<br />
Many of the multiple modeling frameworks considered above, like<br />
IDEF, CIM-OSA, or GERAM for example, have stemmed from integration<br />
efforts in the areas of manufacturing or engineering, and while successful<br />
in their own right, these frameworks are still lagging behind when it comes<br />
to integrating across the multiple dimensions, entities and constituents of a<br />
contemporary enterprise that we have identified above.<br />
7. Epilogue<br />
This paper has provided a description of the main issues on EM and EI.<br />
It has become quite evident from the literature that both the issues are complex<br />
and require knowledge on various disciplines particularly on organizational<br />
management, engineering, and computer science. As such, it can be<br />
generalized that EM and integration mainly revolves around organizational<br />
modeling from which the analysis, design, implementation and integration<br />
of the organization management, business processes, software applications,<br />
and physical computer systems within an enterprise are supported.
134 Nikitas Spiros Koutsoukis<br />
We believe that although we are rich in technology to solve isolated<br />
business problems (e.g., product scheduling, quality control, production<br />
planning, accounting, marketing, etc.), many other problems still reside<br />
within the enterprise. Solving these problems require an integrated approach<br />
to problem solving. This integration must start at the representation level<br />
and the entire enterprise together with its structure, functionality, goals,<br />
objectives, constraints, people, processes, products must be modeled. The<br />
solution will probably be the outcome of an EIM framework.<br />
References<br />
Beeckman, D (1993). CIM-OSA: computer integrated manufacturing open system architecture.<br />
International Journal of Computer Integrated Manufacturing 2(2), 94–105.<br />
Bolton, R, A Dewey, A Goldschmidt and P Horstmann (1997). NIIIP — the national industrial<br />
information infrastructure protocols. In Enterprise Engineering and Integration:<br />
Building International Consensus, K Kosanke and JG Nell (eds.), Berlin: Springer-<br />
Verlag. pp. 293–306.<br />
Bueno, R (1998). Integrated information and process management in manufacturing engineering.<br />
Proc. of the 9th IFAC Symposium on Information Control in Manufacturing<br />
(INCOM’98), pp. 109–112. France: Nancy-Metz, May 24–26.<br />
Chen, D and FVernadat (2002). Enterprise interoperability: a standardisation view. ICEIMT<br />
2002, 273–282.<br />
Christensen, LC, BW Johansen, N Midjo, J Onarheim, TG Syvertsen and T Totland (1995).<br />
Enterprise Modeling-Practices and Perspectives. In Proc. of 9th ASME Engineering<br />
Database Symposium, Rangan (ed.), Boston: Massachusetts, 1171–1181.<br />
Fox, MS (1993). Issues in Enterprise Modeling. Proc. of the IEEE Conference on Systems.<br />
Man and Cybernetics, France: Le Touquet, 83–98.<br />
Fox, MS and M Gruninger (2003). The logic of enterprise modeling, modeling and methodologies<br />
for enterprise integration. In Modelling and Methodologies of Enterprise<br />
Integration, P Bernus and L Nemes (eds.), Cornwall, Great Britain: Chapman and<br />
Hall, 83–98.<br />
Fox, MS and J Huang (2005). Knowledge provenance in enterprise information.<br />
International Journal of Production Research 43(20), 4471–4492.<br />
Giachetti, RE (2004). A framework to review the information integration of the enterprise.<br />
International Journal of Production Research 42(6), 1147–1166.<br />
Hoyte, J (1992). Digital models in engineering — a study on why and how engineers build<br />
and operate digital models for decision support. PHD Thesis, University of Trondheim,<br />
Norway.<br />
Hunt, DV (1991). The Integration of CALS, CE, TQM, PDES, RAMP, and CIM, Enterprise<br />
Integration Sourcebook. San Diego, CA: Academic Press.<br />
Kaplan, RS and DP Norton (1992). The balanced scorecard: measures that drive<br />
performance. Harvard Business Review 70(1), 71–79.<br />
Kemmerer, J (1999). STEP: The Grand Experience, NIST Special Publication SP939,<br />
Washington, C 20402, USA: US Government Printing Office.<br />
Kosanke, K and JG NelI (1997). Enterprise Engineering and Integration: Building<br />
International Consensus. Berlin: Springer-Verlag.
Enterprise Modeling and Integration 135<br />
Kosanke, K, F Vernadat and M Zelm (2000). Enterprise engineering and integration in the<br />
global environment. Advanced Network Enterprises 2000, Berlin, 61–70.<br />
Ladet, P and FB Vernadat (1995). Integrated Manufacturing Systems Engineering. London:<br />
Chapman & Hall.<br />
Lankhorst, M (2005). Enterprise Architecture at Work. Berlin: Springer-Verlag.<br />
Li, H and TJ Williams (2004). A Vision of Enterprise Integration Considerations. Toronto,<br />
Canada: ICEIMT 2004.<br />
Martin, J (1982). Strategic Data-Planning Methodologies. Englewood Cliffs, New York:<br />
Prentice-Hall.<br />
Martin, J, E Robertson and J Springer (2004). Architectural Principles for Enterprise Frameworks:<br />
Guidance for Interoperability. Toronto, Canada: ICEIMT 2004.<br />
Minsky, M (1968). Semantic Information Processing. Cambridge, MA: The MIT Press.<br />
Norrie, M, M Wunderli, R Montau, U Leonhardt, W Schaad and HJ Schek (1995).<br />
Coordination approaches for CIM. In Integrated Manufacturing Systems Engineering.<br />
P Ladet and FB Vernadat (eds.), London: Chapman & Hall pp. 221–235.<br />
Patankar, AK, and S Adiga (1995). Enterprise integration modeling: a review of theory and<br />
practice. Computer Integrated manufacturing Systems 8(1), 21–34.<br />
Petrie, CJ (1992). Enterprise Integration Modeling. Cambridge, MA: The MIT Press.<br />
Porter, M (1985). Competitive Advantage. New York: Free Press.<br />
Prahalad, CK and G Hamel (1990). The core competence of the organisation. Harvard<br />
Business Review 68(3), 79–91.<br />
Rumbaugh, J (1993). Objects in the constitution — enterprise modeling. Journal on Object<br />
Oriented Programming, January issue, 18–24.<br />
Sager, MT (1988). Data concerned enterprise modeling methodologies — a study of practice<br />
and potential. The Australian Computer Journal 20(3), 59–67.<br />
Simon, H (1957). <strong>Models</strong> of Man: Social and Rational. New York: Wiley.<br />
Solvberg, A and DC Kung (1993). Information Systems Engineering — An Introduction.<br />
New York: Springer-Verlag.<br />
Soon, HL, J Neal and A Pennington (1997). Enterprise modeling and integration: taxonomy<br />
of seven key aspects. Computers in Industry 34, 339–359.<br />
Tolone, W, B Chu, G.-J Ahn, R Wilhelm and JE Sims (2002). Challenges to Multi-Enterprise<br />
Integration. ICEIMT 2002, 205–216.<br />
Vernadat, FB (1996). Enterprise Modeling and Integration: Principles and Applications.<br />
London: Chapman & Hall.<br />
Vernadat, F (2002). Enterprise modeling and integration. ICEIMT 2002, 25–33.<br />
Vernadat, FB and BP Zeigler (2000). New simulation requirements for model-based Enterprise<br />
Engineering. Proc. of IFAC/IEEE/INRIA Internal Conference on Manufacturing<br />
Control and Production Logistics (MCPL’2000). Grenoble, 4–7 July 2000. CD-ROM.<br />
Waite, EJ (1998). Advanced Information Technology for Design and Manufacture. In Enterprise<br />
Engineering and Integration: Building International Consensus. K Kosanke and<br />
JG Nell (eds.), Berlin: Springer-Verlag, pp. 256–264.<br />
Williams, TJ (1994). The purdue enterprise reference architecture. Computers in Industry,<br />
24(2–3), 141–158.
This page intentionally left blank
Topic 2<br />
Toward a Theory of Japanese Organizational<br />
Culture and Corporate Performance<br />
Victoria Miroshnik<br />
University of Glasgow, UK<br />
1. Introduction<br />
Japanese national culture has a specific influence on the organizational<br />
culture (OC) of the Japanese companies. In order to understand these aspects<br />
of culture that are affecting corporate performance, a systematic analysis of<br />
the national culture (NC) and OC is essential. The purpose of this chapter<br />
is to provide a scheme to analyze this issue in a systematic way.<br />
National culture can affect corporate OC (Aoki, 1990; Axel, 1995;<br />
Hofstede, 1980) and corporate performance (CP) (Cameron and Quinn,<br />
1999; Kotter and Heskett, 1992). Multinational corporations with its supposedly<br />
common OC demonstrate different behaviors in different countries,<br />
which can affect their performances (Barlett and Yoshihara, 1988; Calori<br />
and Sarnin, 1991). National culture with its various manifestations can create<br />
unique organizational values for a country, which in conjunction with<br />
the human resources management (HRM) practice may create some unique<br />
OC for a corporation. Organizational culture along with a leadership style,<br />
which is the product of OC, NC, and HRM practices, creates a synergy,<br />
which shapes the fortune of the company’s performances. Thus, NC, organizational<br />
values, OCs, leadership styles, human resources practices, and<br />
CPs are inter-related and each depend on the other. If a company maintains<br />
this harmony it can perform well, otherwise it declines.<br />
The structure of this chapter is as follows. In Section 2, relationship<br />
between NCs, OC, and performances, as given in the existing literature, is<br />
analyzed. In Section 3, the Japanese culture was analyzed along with the<br />
controversies on this issue. In Section 4, the new approach of the nationorganization-performance<br />
exchange model (NOPX) is explained in detail<br />
and the theoretical propositions underlying this model are analyzed.<br />
137
138 Victoria Miroshnik<br />
2. Corporate Performance, the Highest Stage of National<br />
and Organizational Culture: A Synthesis<br />
Culture is multidimensional, comprising of several layers of inter-related<br />
variables. As society and organizations are continuously evolving, there is<br />
no theory of culture valid at all times and locations. The core values of a<br />
society (macro-values (MAVs)) can be analyzed by asking the following<br />
questions:<br />
(a) What are the relationships between human and nature? (b) What are<br />
the characteristics of innate human nature? (c) What is the focus regarding<br />
time — whether past, future, or present, or a combination of all these?<br />
(d) What are the modalities of human activity, whether spontaneous or<br />
introspective or result-oriented or a combination of these? (e) What is the<br />
basis of relationship between one man and another in the society?<br />
In the structure of the NC, there are certain micro or individual values,<br />
which include sense of belonging, excitement, fun and enjoyment,<br />
warm relationship with others, self-fulfilment, being well-respected, and a<br />
sense of accomplishment, security, and self-respect (Kahle, et al., 1988).<br />
Although Hofstede (1990; 1993) has the opinion that the perceived practices<br />
are the roots of the OC rather than values, most researchers have accepted<br />
the importance of values in shaping cultures in several organizations.<br />
Combinations of micro-values (MIVs) and macro-values (MAVs) can<br />
give rise to an OC, given certain meso-values (MEV) or organizational<br />
values, which are specific for a country. Meso-values are certain codes of<br />
behavior, which influences the future, and the expected behavior of the<br />
members of the society, if they want to belong to the main stream. These<br />
vary from one society to another, as the expectations of different societies,<br />
as products of historical experiences, religious, and moral values are<br />
different.<br />
Cultural values are shared ideas about what is good, desirable, and<br />
justified; these are expressed in a variety of ways. Social order, respect for<br />
traditions, security, and wisdom are important for the society, which is like<br />
an extended family.<br />
According to Schein (1997), OC emerges from some common assumptions<br />
about the organization, which the members share as a result of their<br />
experiences in that organization. These are reflected in their pattern of<br />
behaviors, expressed values, and observed artefacts. Organizational culture<br />
provides accepted solutions to known problems, which the members<br />
learn and feel about and forms a set of shared philosophies, expectations,
Toward a Theory of Japanese Organizational Culture 139<br />
norms, and behavior patterns, which promotes higher level of achievements<br />
(Kilman et al., 1985; Marcoulides and Heck, 1993; Schein, 1997).<br />
Organizational culture, for a large and geographically dispersed organization,<br />
may have many different cultures. According to Kotter and Heskett<br />
(1992), leaders create certain vision or philosophy and business strategy<br />
for the company. A corporate culture emerges that reflects the vision and<br />
strategy of the leaders and experiences they had while implementing these<br />
strategies. However, the question is whether it is valid for every NC.<br />
A combination of MAVs, MEVs, and MIVs creates a specific OC,<br />
which varies from country to country according to their differences in<br />
NC. Hofstede (1980; 1985; 1990; 1993) showed the relationship between<br />
NC and OC. Global leadership and organizational behavior effectiveness<br />
(GLOBE) research project has tried to identify the relationships among<br />
leadership, social culture, and OC (House, 1999).<br />
Cameron and Quinn (1999) have mentioned that the most important<br />
competitive advantage of a company is its OC. If an organization<br />
has a “strong culture” with “well-integrated and an effective” set of values,<br />
beliefs, and behaviors, it normally demonstrates high level of CPs<br />
(Ouchi, 1981).<br />
Denison and Mishra (1995) have attempted to relate OC and performances<br />
based on different characteristics of the OC. However, different<br />
types of OC enhance different types of business (Kotter and Heskett, 1992).<br />
There is no single cultural formula for long-run effectiveness. As Siehl and<br />
Martin (1988) have observed, culture may serve as a filter for factors that<br />
influence the performance of an organization. These factors are different for<br />
different organizations. Thus, a thorough analysis regarding the relationship<br />
between culture and performances is essential.<br />
3. Japanese National and Organizational Culture<br />
Japanese firms are considered to be of a family unit with long-term orientations<br />
for HRM and with close ties with other like-minded firms,<br />
banks, and the government. The OC promotes loyalty, harmony, hard<br />
work, self-sacrifice, and consensus decision-making. These, along with<br />
lifetime employment and seniority-based promotions, are considered to<br />
be the natural outcome of the Japanese NC. (Ikeda, 1987; Imai, 1986;<br />
Ouchi, 1981). Japanese workers’ psychological dependency on companies<br />
emerges from their intimate dependent relationships with the society and the<br />
nation.
140 Victoria Miroshnik<br />
Japanese NC has certain MIVs: demonstration of appropriate attitudes<br />
(taido); the way of thinking (kangaekata); and spirit (ishiki), which forms<br />
the basic value system for the Japanese (Kobayashi, 1980). Certain MEVs<br />
can be identified as the core of the cultural life for the Japanese, if they want<br />
to belong to the mainstream Japanese society (Basu, 1999; Fujino, 1998).<br />
These are (a) Senpai-Kohai system; (b) Conformity; (c) Hou-Ren-Sou and<br />
(d) Kaizen or continuous improvements.<br />
Senpai-Kohai or senior-junior relationships are formed from the level<br />
of primary schools where junior students have followed the orders of the<br />
senior students, who in turn, may help the juniors in learning. The process<br />
continues throughout the lifetime for the Japanese. In work places, Senpais<br />
will explain Kohais how to do their work, the basic code of conducts and<br />
norms.<br />
From this system emerges the second MEV, Conformity, which is better<br />
understood from the saying that “nails that sticks out should be beaten<br />
down”. The inner meaning is that unless someone conforms to the rules<br />
of the community or co-workers, he/she would be an outcaste. There is no<br />
room for individualism in the Japanese society or in work place.<br />
The third item, Hou-Ren-Sou, is the basic feature of the Japanese<br />
organizations. Hou-Ren-Sou, is a combination of three different words in<br />
Japanese: Houkoku i.e., to report, Renraku i.e., to inform and Soudan i.e.,<br />
to consult or pre-consult.<br />
Subordinates should always report to the superior. Superiors and subordinates<br />
share information. Consultations and pre-consultations are required;<br />
no one can make his/her decision by himself/herself even within the delegated<br />
authority. There is no space in which the delegation of authority may<br />
function. Combining these words, Houkoku, Renraku, and Soudan, “Hou-<br />
Ren-Sou” is the core value of the culture in Japan. Making suggestions, for<br />
improvements, without pre-consultations is considered to be an offensive<br />
behavior in Japanese culture. Everyone from the clerk to the president, from<br />
the entry day of the working life to the date of retirement, every Japanese<br />
must follow the “Hou-Ren-Sou” value system.<br />
Kaizen i.e., continuous improvement is another MEV that is one of the<br />
basic ingredients of the Japanese culture. Search for continuous improvement<br />
during the Meiji Government of the 19th century led them to search the<br />
world for knowledge. Japanese companies today are doing the same in terms<br />
of both acquisitions of knowledge by setting up R&D centers throughout<br />
the western world, and by having “quality circles” in work places in order to<br />
implement new knowledge to improve the product quality and to increase<br />
its efficiency (Kujawa, 1979; Kumagai, 1996; Miyajima, 1996).
Toward a Theory of Japanese Organizational Culture 141<br />
3.1. Japanese OC<br />
In order to understand the Japanese OC, it is essential to know the Japanese<br />
system of management, which gave rise to the unique Japanese style of OC.<br />
What is written below is the result of several visits to the major Japanese<br />
corporations (like Toyota, Mitsubishi, Honda, and Nissan) and interviews<br />
with several senior executives, including members of the board and vicepresidents<br />
of these companies.<br />
Japanese system of management is a complete philosophy of organization,<br />
which can affect every part of the enterprise. There are three basic<br />
ingredients: lean production system, total quality management (TQM) and<br />
HRMs (Kobayashi, 1980). These three ingredients are inter-linked in order<br />
to produce total effect on the management of the Japanese enterprises.<br />
Because Japanese overseas affiliates are part of the family of the parent<br />
company, their management systems are part of the management strategy<br />
of the parent company (Basu, 1999; Morishima, 1996; Morita, 1992;<br />
Shimada, 1993).<br />
The basic idea of the lean production system and the fundamental<br />
organizational principles is the “Human-ware” (Shimada, 1993), which is<br />
described in Fig. 1. “Human-ware” is defined as the integration and interdependence<br />
of machinery and human relations and a concept to differentiate<br />
among different types of production systems. The purpose of the lean production<br />
philosophy, which was developed at the Toyota Motor Company,<br />
is to lower the costs. This is done through the elimination of waste —<br />
everything that does not add value to the product. The constant strive for<br />
perfection (Kaizen in Japanese) is the overriding concept behind good management,<br />
in which the production system is being constantly improved;<br />
perfection is the only goal. Involving everyone in the work of improvement<br />
is often accomplished through quality circles. Lean production system uses<br />
“autonomous defect control” (Pokayoke in Japanese), which is an inexpensive<br />
means of conducting inspection for all units to ensure zero defects.<br />
Quality assurance is the responsibility of everyone. Manufacturing tasks<br />
are organized into teams. The principle of “just-in-time” (JIT) means, each<br />
NC OC<br />
LC CP<br />
Where “NC” is “National Culture”; “OC” is “Organizational Culture”; “LC” is<br />
“Leader Culture”; and, “CP” is “Corporate Performance”.<br />
Figure 1.<br />
Sub-model 1: culture-performance system.
142 Victoria Miroshnik<br />
process should be provided with the right part, in the right quantity at exactly<br />
the right point of time. The ultimate goal is that every process should be<br />
provided with one part at a time, exactly when that part is needed.<br />
The most important feature of the organizational set-up of the lean<br />
production system is the extensive use of multifunctional teams, which<br />
are groups of workers who are able to perform many different works. The<br />
teams are organized along a cell-based production flow system. The number<br />
of job-classifications also declines. Workers have received training to<br />
perform a number of different tasks, such as the statistical process control,<br />
quality instruments, computers, set-up performances, maintenance etc. In<br />
the lean production system responsibilities are decentralized. There is no<br />
supervisory level in the hierarchy. The multifunctional team is expected<br />
to perform supervisory tasks. This is done through the rotations of team<br />
leadership among workers. As a result, the number of hierarchical levels in<br />
the organization can be reduced. The number of functional areas that are<br />
the responsibility of the teams increases (Kumazawa and Yamada, 1989).<br />
In a multifunctional set-up, it is vital to provide information in time<br />
and continuously in the production flow. The production system is changing<br />
and gradually adopting a more flexible system in order to adopt itself to<br />
changes in demand; which may reduce costs of production and increase efficiency.<br />
There are extensive usages of Kaizen activities and TQM and Total<br />
Productive Maintenance (TPM) to increase the effectiveness of corporate<br />
performances.<br />
4. A Nation-Organization-Leader-Performance Model<br />
Organizational culture is a product of a number of factors: NC, human<br />
resources practice, and leadership styles have prominent influences on it.<br />
These in turn, along with the OC, affect CPs. The proposed model, as in<br />
Fig. 1, is the synthesis of the ideas relating to this central theme. Figures 2<br />
and 3 describe the proposed theoretical model relating OC and CPs. Corporate<br />
performances are measured in terms of the following criteria:<br />
(a) customers’ satisfaction;<br />
(b) employee’s satisfaction and commitment and<br />
(c) contributions of the organization to the society and environment.<br />
Most companies in Japan use these criteria to signify CPs (Basu, 1999;<br />
Nakane, 1970).<br />
There are several latent or hypothetical variables, which jointly can<br />
influence the outcome. National culture is affected by two constructs:
Toward a Theory of Japanese Organizational Culture 143<br />
MIVs and MAVs. Micro-values (MIVs) are composed of several factors<br />
(Kahle et al., 1988). These are: (1) the sense of belonging; (2) respect<br />
and recognition from others and (3) a sense of life accomplishments and<br />
(4) self-respect.<br />
Macro-values, according to the opinions of the executives of the<br />
Japanese, firms are religious values, habitual values, and moral values.<br />
There may be other MAVs, which are important for other nations like geography,<br />
racial origin, etc., but for the Japanese, these are not so important<br />
(Basu, 1999). Although moral and habitual values can be the results of<br />
religious values, it is better to separate out these three values as distinct.<br />
The moral and habitual values are different in Japan from other East Asian<br />
nations. Honor and “respect from others” are central to the Japanese psychology.<br />
Ritual suicides (Harakiri) are honorable acts for the Japanese, if<br />
they fail in some way to do their duty. Habitual values are extreme politeness<br />
on the surface, beautifications of everything, community spirit, and cleanness<br />
are unique in Japan, which are not followed in any other countries<br />
in Asia.<br />
This is particularly true if we examine certain MEVs, which are neither<br />
MIV nor MAV values, but are derived from the NC. It is possible to identify<br />
five different MEVs, which are important outcomes of the Japanese NC.<br />
These are discussed as follows:<br />
(a) Exclusivity or insider-outsider (Uchi-Soto in Japanese) psychology by<br />
which Japanese exclude anyone who is not an ethnic Japanese from<br />
social discourse; (It is different from color or religious exclusivities. For<br />
example, the Chinese or Koreans, who are living in Japan for centuries,<br />
are excluded from the Japanese social circles.)<br />
(b) Conformity or the doctrine of “nail that sticks up should be beaten<br />
down”(Deru Kuiwa Utareru in Japanese) — deviations from the mainstream<br />
norms are not tolerated;<br />
(c) Seniority system (Senpai-Kohai in Japanese) by which every junior<br />
must obey and show respects to the seniors;<br />
(d) Collectivism in decision-making process (“Hou-Ren-Sou” system in<br />
Japanese) and<br />
(e) Continuous improvements (Keizen in Japanese), which is the fundamental<br />
philosophy of the Japanese society.<br />
These MEVs are exclusively Japanese, a reflection of the unique<br />
Japanese culture (Basu, 1999; Nakane, 1970) and is fundamental to the<br />
Japanese organizations.
144 Victoria Miroshnik<br />
National culture construct along with the MEVs affect both the organizational<br />
culture construct (OR), HRM system and leadership styles (LS)<br />
constructs. Japanese OR construct is affected by the NC, MEVs, and the<br />
HRM system. The similarities in OCs in different companies are due to<br />
the similarities in NC and MEV; the differences are due mainly to the differences<br />
in HRM. In Japan human resource practices vary from one company<br />
to another and as a result, OCs and performances vary. In Toyota,<br />
it is very consistent and strong culture with great emphasis on discipline,<br />
Keizen, TQM and just in time (JIT) production-inventory system.<br />
In many other companies in Japan, situations are not the same; as a result,<br />
OC differs.<br />
Organizational culture is affected by the NC, MEVs, and HRM. If the<br />
company’s leader is also the founder, then only the leadership style (LS) can<br />
affect the HRM and OC. Otherwise, when the leaders are professional managers<br />
trained and grown up within the company (In Japan and in Japanese<br />
companies abroad, it is out of practice to accept an outsider for the executive<br />
positions. All executives enter the company after their graduation and<br />
stay until they retire), appropriate LCs are developed by the OC and HRM.<br />
Due to the collective decision-making process (Hou-Ren-Sou) leadership,<br />
a MEV in the model, cannot affect an organizational culture or the<br />
HRM system. It is very different from the Western (America, European,<br />
or Australian) companies where the leaders are hired from outside and are<br />
expected to be innovative regarding the OC and HRM system. This is quite<br />
alien to the Japanese culture. Leadership style in Japan is an outcome, not<br />
an independent variable as in the Western companies. Leadership style is<br />
affected by the NC, MEVs, OC, and the HRM.<br />
Organizational culture (OC) is affected by four factors, according to this<br />
proposed model. These factors are: (a) stability; (b) flexibility; (c) internal<br />
focus and (d) external focus (Cameron and Quinn, 1999). These factors are<br />
universal, not restricted to the Japanese companies. However, how an OC<br />
is being affected by NC, MEVs, and HRM would vary from country to<br />
country.<br />
Leadership style is a function of four factors. These factors are leader:<br />
(a) as a facilitator; (b) as an innovator; (c) as a technical expert and (d) as<br />
an effective competitor for rival firms (Cameron and Quinn, 1999). These<br />
four characteristics of a leader are universal, but the emphasis varies from<br />
country to country.<br />
Finally, the corporate performance (CP) is affected by a number of<br />
influencing factors: (a) customers’ satisfaction, (b) employees’ satisfaction
Toward a Theory of Japanese Organizational Culture 145<br />
MAV<br />
LC<br />
CP<br />
NC<br />
OC<br />
MEV<br />
MIV<br />
HRMPJ<br />
Where “NC” is “National Culture”; “OC” is “Organizational Culture”; “LC” is “Leader<br />
Culture”; “MAV” is “Macro-Values of Culture”; “MEV” is “Meso-Values of Culture”;<br />
“MIV” is “Micro-Values of Culture”; “HRMPJ” is “Human Resources Management<br />
Practices in Japan”, and “CP” is “Corporate Performance”.<br />
Figure 2.<br />
Value–culture–performance: a thematic presentation.<br />
and commitments, and (c) the contribution of the organization to the society.<br />
If we consider the contribution of the organization to society that is<br />
already taken into account by the customers’ satisfaction and employees’<br />
satisfaction, then CP has these two important contributory factors.<br />
Theoretical relationship between the observed variables and the underlying<br />
constructs are specified in an exploratory model, represented in Fig. 2.<br />
In Fig. 3, a more elaborate model, including the unique HRM system of the<br />
Japanese companies is described. Figure 3 represents a Linear Structural<br />
Relations (LISREL) version (Joreskog and Sorbom, 1999) of the model,<br />
which can be estimated using the methods of structural equation modeling<br />
(Raykov and Marcoulides, 2000).<br />
5. Discussion<br />
The proposed theory describes the relationship between NC and CPs<br />
through OC with the HRM system affecting the nature of the culture and<br />
leadership. These inter-relationships are important aspects of the Japanese<br />
corporate behaviors in a multinational setting and this proposed theory is<br />
an attempt to understand the Japanese corporate culture.<br />
Analysis of the Japanese culture and organizations by outsiders so far<br />
suffers from a number of defects (Ikeda, 1987; Watanabe, 1998). They neither<br />
try to understand the effects of the Japanese value system on their<br />
organizations nor do they give any importance to the relationship between
146 Victoria Miroshnik<br />
A1<br />
A2<br />
A3<br />
D1<br />
D2 D3 D4<br />
B1<br />
B2<br />
B3<br />
MAV<br />
NC OC<br />
LC CP<br />
H1<br />
B4<br />
MEV<br />
MIV<br />
G1 G2 G3 FG<br />
F6<br />
H21<br />
C1 C2 C3 C4 C5<br />
HRMPJ<br />
F1<br />
F2<br />
F3<br />
F4<br />
F5<br />
E1<br />
E24<br />
Where “NC” is “National Culture”; “OC” is “Organizational Culture”; “LC” is “Leader Culture”;<br />
“MAV” is “Macro-Values of Culture”; “MEV” is “Meso-Values of Culture”; “MIV” is<br />
“Micro-Values of Culture”, and “CP” is “Corporate Performance”; A1 “Religious Values”, A2<br />
“Moral Values” A3 “Habitual Values” are indicators for MAV variable; B1 “Power Distance”,<br />
B2 “Individualism/Collectivism”, B3 “Masculinity/Femininity” are indicators for NC variable;<br />
C1 “Uchi/Soto Value”, C2 “Deru Kuiwa Utareru Value”, C3 “Senpai/Kohai Value”, C4<br />
“Hou-Ren-Sou Value”, C5 “Kaizen Value” are indicators for MEV variable; D1 “Stability”,<br />
D2 “Flexibility”, D3 “External Focus”, D4 “Internal Focus are indicators for OC variable”;<br />
E1 “On-Job-Training”; E2 “Emphasis on Personality of Recruits” are indicators for HRMPJ<br />
variable; F1 “Sense of Belonging to a Group Value”, F2 “Respect and Recognition from Others<br />
Value”, F3 “Sense of Life Accomplishment Value”, F4 “Self-respect Value”, F5 “Kangaekata<br />
Value”, F6 “Ishiki Value” are indicators for MIV variable; G1 “Leader as Facilitator”, G2<br />
“Leader as Innovator”, G3 “Leader as Technical Expert”, G4 “Leader as Competitor” are<br />
indicators for LC variable; H1 “Customers Satisfaction”, H2 “Employees satisfaction and<br />
Commitment” are indicators for CP variable.<br />
Figure 3. Research model: Values-culture-performance (VCP): predicted paths<br />
between variables and indicators.<br />
performance and culture. So far, the success of the Japanese companies was<br />
analyzed only in terms of their unique production and operations management<br />
system. However, the inter-relationship between the production and<br />
operations management system, which is a part of the OC, and the specific<br />
HRM system it requires for its proper implementation, was not given<br />
sufficient attention.<br />
The typical example is the analysis of the failure of the implementations<br />
of Japanese operations management system in Britain by Oliver and<br />
Williamson (1992) or in China by Taylor (1999). They could not explain<br />
the reason for the failure, as they have not analyzed the Japanese OC and<br />
particularly the HRM system.
Toward a Theory of Japanese Organizational Culture 147<br />
In a Japanese company, the LSs and the OC are designed by the HRM<br />
system; these are not coming from outside. An effective corporate performance<br />
is the result of these underlying determinants of OC and LSs. Thus,<br />
in a Japanese organization, LS is rooted in the HRM system, which emerges<br />
from the MEVs of the Japanese national culture.<br />
Whether efficient CPs can be repeated in foreign locations depends on<br />
the transmission mechanism of the Japanese OC, which in turn, depends on<br />
the adaptations of the Japanese style HRM system. This can face obstacles<br />
due to the different MEVs in a foreign society and as a result, OC in a<br />
foreign location may be different from the OC of the company in Japan.<br />
Adaptation of the operations management system alone will not create an<br />
effective performance of a Japanese company in a foreign location. Creation<br />
of an appropriate HRM system and appropriate OC are essential for an<br />
effective performance.<br />
6. Conclusion<br />
A proposed theory of the relationship, among NC, OC, human resources<br />
practices, and CPs for the Japanese multinational companies, is narrated.<br />
Cultural differences are important and these factors affect every component<br />
of the company behaviors and performances. The question was raised<br />
regarding the factors, which are relatively more important than the others<br />
and whether it will be possible to manipulate these factors in order to<br />
increase the performances of the multinational companies. The proposed<br />
theory may provide an answer to these issues.<br />
References<br />
Aoki, M (1990). Towards an economic model of the Japanese firm. Journal of <strong>Economic</strong><br />
Literature 28, 1–27.<br />
Axel, M (1995). Culture-bound aspects of Japanese management. Management International<br />
Review 35(2), 57–73.<br />
Barlett, CA and H Yoshihara (1988). New challenges for Japanese multinationals: is organization<br />
adaptation their Achilles Heel. Human Resources Management 27(1), 19–43.<br />
Basu, S (1999). Corporate Purpose. New York: Garland Publishers.<br />
Calori, R and P Sarnin (1991). Corporate culture and economic performance: a French study.<br />
Organization Studies 12(1), 49–74.<br />
Cameron, KS and R Quinn (1999). Diagnosing and Changing Organizational Culture:<br />
Based on the Competing Values Framework. New York: Addison-Wesley.<br />
Denison, DR and AK Mishra (1995). Toward a theory of organizational culture and effectiveness.<br />
Organizational Science 6(2), March–April, 204–223.
148 Victoria Miroshnik<br />
Fujino, T (1998). How Japanese companies have globalized. Management Japan 31(2),<br />
1–28.<br />
Hofstede, G (1980). Culture’s Consequences: International Differences in Work-related<br />
Values. Beverly Hills, CA: Sage.<br />
Hofstede, G (1985). The interaction between national and organisational value system.<br />
Journal of Management Studies 22(3), 347–357.<br />
Hofstede, G (1990). Measuring organizational cultures: a qualitative and quantitative study<br />
across twenty cases. Administrative Science Quarterly 35, 286–316.<br />
Hofstede, G (1993). Cultural constraints in management theories. Academy of Management<br />
Executive 7(1), 81–94.<br />
House, RJ et al. (1999). Cultural influences on leadership: Project GLOBE. In Advances in<br />
Global Leadership, W Moblet, J Gessner and V Arnold (eds.), Vol. 1. Greenwick: CT,<br />
JAI Press, 171–234.<br />
Ikeda, K (1987). Nihon oyobi nihonjin ni tsuiteno imeji (Image of Japan and Japanese). In<br />
Sekai wa Nihon wo Dou Miteiruka (How Does the World See Japan). A Tsujimura,<br />
K Furuhata and H Akuto (eds.) Tokyo: Nihon Hyoronsa, pp. 12–31.<br />
Imai, M (1986). Kaizen: The Key to Japan’s Competitive Success. New York, NY:<br />
McGraw-Hill.<br />
Joreskog, KG and D Sorbom (1999). LISREL 8.30: User’s Reference Guide. Chicago:<br />
Scientific Software.<br />
Kahle, LR, RJ Best and RJ Kennedy (1988). An alternative method for measuring valuebased<br />
segmentation and advertisement positioning. Current Issues in Research in<br />
Advertising 11, 139–155.<br />
Kilman, R, MJ Saxton, and R Serpa (1985). Introduction: five key issues in understanding<br />
and changing culture, In Gaining Control of the Organizational Culture, R Kilman,<br />
M Gaxton and R Serpa (eds.), San Francisco: Jossey-Bass, pp. 1–16.<br />
Kobayashi, N (1980). Nihon no Takokuseki Kigyo (Japanese multinational corporations).<br />
Tokyo: Chuo Keizaisha.<br />
Kotter, JP and JL Heskett (1992). Corporate Culture and Performance. NY: The Free Press.<br />
Kujawa, D (1979). The labour relations of US multinationals abroad: comparative and<br />
prospective views. Labour and Society, 4, 3–25.<br />
Kumagai, F (1996). Nihon Teki Seisan Sisutemu in USA (Japanese Production System in<br />
USA). Tokyo: JETRO.<br />
Kumazawa, M and J Yamada (1989). Jobs and skills under the lifelong Nenko employment<br />
practice. In The Transformation of Work. S Wood (ed.). London: Unwin Hyman, 56–79.<br />
Marcoulides, GA and RH Heck (1993). Organization culture and performance: proposing<br />
and testing a model. Organization Science 14(2), May, 209–225.<br />
Morishima, M (1996). Renegotiating psychological controls, Japanese style. In Trends in<br />
Organizational Behavior. CL Cooper (ed.), Vol. 3. NewYork: John Wiley & Sons Ltd.,<br />
149–165.<br />
Morita, A (1992). A moment for Japanese management. Japan-Echo 19(2), 1–12.<br />
Miyajima, H (1996). The evolution and change of contingent governance structure in the<br />
J-firm system: An approach to presidential turnover and firm performance. Working<br />
Paper No. 9606. Institute for Research in Contemporary Political and <strong>Economic</strong>Affairs.<br />
Tokyo: Waseda University.<br />
Nakane, G (1970). Japanese Society. Harmondsworth: Penguin.
Toward a Theory of Japanese Organizational Culture 149<br />
Oliver, N and B Wilkinson (1992). The Japanization of British Industry. Cambridge,<br />
Massachussets: Blackwell.<br />
Ouchi, W (1981). Theory Z: How American Business can Meet the Japanese Challenge.<br />
Reading, MA: Addison-Wesley.<br />
Raykov, T and GA Marcoulides, (2000). A First Course in Structural Equation Modeling.<br />
London: Lawrence Erlbaum.<br />
Schein, EH (1997). Organizational Culture and Leadership. San-Francisco: Jossey-Bass<br />
Publishers.<br />
Shimada, H (1993). Japanese management of auto-production in the United States: An<br />
overview of human technology in international labour organization. In Lean Production<br />
and Beyond, Geneva: ILO, 23–47.<br />
Taylor, B (1999). Patterns of control within Japanese manufacturing plants in China: doubts<br />
about Japanization in Asia. Journal of Management Studies 36(6), 853–873.<br />
Watanabe, S (1998). Manager-subordinate relationship in cross-cultural organizations: the<br />
case of Japanese subsidiaries in the United States. International Journal of Japanese<br />
Sociology 7, 23–43.
This page intentionally left blank
Topic 3<br />
Health Service Management Using<br />
Goal Programming<br />
Anna-Maria Mouza<br />
Institute of Technology and Education, Greece<br />
1. Introduction<br />
Private production units usually operate on the basis of profit maximization.<br />
However, to face competition from similar units and to be in line with some<br />
major socio-economic factors, some decision makers are forced to relax this<br />
basic objective. In order to present propositions for optimal management<br />
decisions, one should combine the priorities of the decision maker with the<br />
technical and socio-economic factors, which are involved in the operating<br />
process in the best possible way. The production unit considered in this<br />
paper is a relatively small (60 beds) private clinic in northern Greece, which<br />
is similar to the orthopedic department of a hospital studied and described<br />
elsewhere (Mouza, 1996).<br />
However, in this particular case, there is no out-patient services. To<br />
facilitate the presentation, I consider a five-year operational plan where<br />
the personnel claims, the manpower working pattern, the various expenses,<br />
together with the profit maximization target, and the expected number of<br />
patients are described in detail. The necessary projections are based on<br />
reliable techniques presented elsewhere (Mouza, 2002; 2006).<br />
After setting the priorities of the various targets, I formulate a proper<br />
“goal programming” problem using deviational variables (di<br />
− and d<br />
i + ) and<br />
obtained the first-run solution. Then, I proceed to the goal attainment evaluation,<br />
which dictated a re-adjustment of the priorities. The second-run<br />
solution indicates that a slower acceleration of the profit maximization rate<br />
is preferable, to avoid over-charging the patients, at the same time satisfying<br />
most of the requirements for personnel management. Furthermore, if the<br />
decision maker’s preference is relatively rigid, a lower rate of the increase<br />
151
152 Anna-Maria Mouza<br />
of salaries of the employees should be adopted. All computational results<br />
are presented at the end of the relevant section.<br />
2. The Details of the Private Clinic in Relation<br />
to the Five-Year Plan<br />
In Table 1, I present the composition of the personnel, and their salary<br />
at the base year (t 0 ). Note that the 5-year planning horizon, refers to the<br />
time-periods (years) t 1 , t 2 , t 3 , t 4 , and t 5 .<br />
In columns 4 and 5 of Table 1, the total amount of the wages for all<br />
employees of the same group, for the initial and final year is stated. Note<br />
that the wages are an average of the salaries earned by each person in<br />
the individual group. For each group, an average percent of annual salary<br />
Table 1.<br />
Personnel of the orthopedic clinic and their wages.<br />
Groups<br />
Position of the<br />
group members<br />
Members<br />
at each<br />
group<br />
Annual salary at<br />
year t 0 for all<br />
group members<br />
(in thousand )<br />
Salary at final<br />
year t 5 , for all<br />
group members<br />
(in thousand )<br />
1 Orthopedic surgeons, 10 260 356.222<br />
anaesthiologists and<br />
microbiologists<br />
2 Pharmacist 1 24.7 33.054<br />
3 Operating-room nurses 3 54.6 68.04<br />
4 Nurses plus laboratory 14 163.8 199.29<br />
staff<br />
5 Axial-tomography and 3 58.5 74.307<br />
X-ray technicians<br />
6 Clinic manager 1 27.3 36.190<br />
7 Secretary and<br />
3 32.76 39.476<br />
administrative<br />
service staff<br />
8 Office clerks 4 41.6 49.647<br />
9 Cooking and food 4 42.64 50.643<br />
service staff<br />
10 Other staff (guard,<br />
bearers maintenance,<br />
cleaning service etc.)<br />
10 101.4 119.851
Health Service Management 153<br />
increase, say R i , is considered. It should be noted that in some groups, this<br />
rate is well above the rate of inflation growth, and the rate stated in the<br />
collective wage and salary agreements for each employees group. These<br />
rates of annual salary increase for all groups are presented in Table 3.<br />
Denoting by A 0 (i) the annual salary of all members of group i at the base<br />
year t 0 , and the final annual salary at year t 5 for all members of the same<br />
group, denoted by A n (i), is computed from:<br />
(<br />
A n (i) = A 0 (i) 1 + R ) 5<br />
i<br />
. (1)<br />
100<br />
Thus for group 1, where R 1 = 6.5%, the final salary for all group members<br />
will be:<br />
A n (1) = 260(1 + 0.065) 5 = 356.222.<br />
It should be noted, that these rates of salary increases have been formed<br />
in accordance to the personnel claims. Other additional expenses for the<br />
final year of of the planning period are presented in Table 2.<br />
On the top of Table 2, the forecast regarding the expected number<br />
of patients for the final year is stated. It should be pointed out that not<br />
all patients necessarily need an axial-tomography and/or X-ray treatment.<br />
That is why the average per patient cost stated in Table 2 seems to be<br />
relatively low.<br />
The interactive voice response (IVR) systems which are effectively used<br />
in healthcare and hospitals are described by Mouza (2003). It should be<br />
noted that replacement expenses together with expenses for obtaining new<br />
devices, seems more reasonable to be equally distributed over the five years<br />
of the planning period. Needless to say, that in case of replacements, the<br />
estimated salvage on old equipment has to be subtracted from the relevant<br />
cost. Thus, at the final year, only one fifth of such expenses is stated. For the<br />
case under consideration, the replacement refers to the server of the local<br />
area computer network (LAN), together with two stations. Their net cost<br />
sum up to 2000, which has been equally spread over the five years of the<br />
planning period.<br />
From Table 1, I computed the average salary per employee at the final<br />
year which, when multiplied by the group size gives the total salary for the<br />
corresponding group, as it can be verified from Table 3.<br />
It can be seen from Table 3, that X i (i = 1,...,10) denotes the size (i.e.,<br />
the members) of group i and c i , the average salary of each group member.<br />
Thus, the total salary of all the members of group 1, for instance, can also
154 Anna-Maria Mouza<br />
Table 2.<br />
final year.<br />
Forecasted number of patients and various expenses for the<br />
Patient expenses<br />
Forecasted average<br />
per patient, at the final<br />
period t f ( )<br />
Forecasted number of patients for the period (year) t 5 = 3625<br />
Axial-tomography 9.1<br />
X-ray 5.7<br />
Medical supplies 59.4<br />
Administrative and<br />
32.1<br />
miscellaneous<br />
Other expected<br />
Total ( ) Amount for the final year t 5<br />
expenses<br />
Installation of IVR<br />
3000 600<br />
system<br />
Server and computers 2350 (salvage value: 350) 400<br />
replacement<br />
Personnel insurance<br />
(19% of total<br />
Endogenous (decision<br />
variable X 39 )<br />
salaries)<br />
Laboratory expenses 60,000<br />
Fixed expenses plus<br />
180,000<br />
depreciation<br />
All other expenses 80,000<br />
be computed from:<br />
c 1 × X 1 = 35622.2 × 10 = 356222.<br />
If I denote by X 10+i , the salary expenses of group i, then I may write<br />
X 10+i = c i × X i (i = 1,...,10). (2)<br />
so that the total salary expenses for the final year can be determined by<br />
evaluating the summation:<br />
20∑<br />
i=11<br />
X i . (3)
Health Service Management 155<br />
Table 3.<br />
Employees’salary at the initial and final year of the planning period.<br />
Groups (i)<br />
Number<br />
of group<br />
members<br />
(X i )<br />
Annual<br />
salary at<br />
year t 0 ,for<br />
all the<br />
members of<br />
group i (in<br />
thousand )<br />
Annual<br />
rate of<br />
increase,<br />
R i (in %)<br />
R i<br />
Salary at<br />
final year t f ,<br />
for all the<br />
members of<br />
group i<br />
(in )<br />
Average<br />
per person<br />
at each<br />
group (c c i )<br />
1 10 260 6.5 356,222 35,622.2<br />
2 1 24.7 6 33,054 33,054<br />
3 3 54.6 4.5 68,040 22,680<br />
4 14 163.8 4 199,290 14,235<br />
5 3 58.5 4.9 74,307 24,769<br />
6 1 27.3 5.8 36,190 36,190<br />
7 3 32.76 3.8 39,476 13,158.666<br />
8 4 41.6 3.6 49,647 12,411.75<br />
9 4 42.64 3.5 50,643 12,660.75<br />
10 10 101.4 3.4 119,851 11,985.10<br />
Another point of interest refers to the average working hours of the<br />
members of each group. The relative figures are analytically presented in<br />
Table 4.<br />
It can be easily verified that in Table 4, column 5 is obtained by multiplying<br />
the elements of column 4 by 52.<br />
Denoting by X 22+i (i = 1,...,10) the total average working hours per<br />
year for all the group members, then I may write<br />
X 22+i = w i × X i (i = 1,...,10). (4)<br />
3. The Nature of the Problem<br />
Given all the information presented in Tables 1–4, the resultant problem<br />
is, how to formulate a feasible operating plan, which should combine<br />
the desired salary increases and the full utilization of working hours,<br />
the expenses incurred, together with the desired gross profit at the final<br />
year set by the decision maker, in the best possible way with the minimum<br />
sacrifice. This problem can be answered by the goal programming<br />
method, originally developed by Charnes and Cooper (1961). It should
156 Anna-Maria Mouza<br />
Table 4.<br />
Working hours of each group personnel.<br />
Groups (i)<br />
Number<br />
of group<br />
members<br />
Hours per<br />
week<br />
(average),<br />
for each<br />
member of<br />
group i (in<br />
thousand )<br />
Total hours<br />
per week<br />
for all the<br />
members of<br />
group i<br />
Total<br />
hours per<br />
year, for<br />
all the<br />
members<br />
of each<br />
group<br />
Average<br />
per person<br />
at each<br />
group (w i )<br />
1 10 65 650 33,800 3380<br />
2 1 45 45 2340 2340<br />
3 3 65 195 10,140 3380<br />
4 14 40 560 29,120 2080<br />
5 3 40 120 6240 2080<br />
6 1 45 45 2340 2340<br />
7 3 40 120 6240 2080<br />
8 4 40 160 8320 2080<br />
9 4 30 120 6240 1560<br />
10 10 30 300 15,600 1560<br />
be recalled that goal programming is a mathematical programming technique<br />
with the capability of handling multiple goals given certain priority<br />
structure. Furthermore, a goal programming model may be composed of<br />
non-homogeneous units of measure, which is the case under consideration.<br />
In brief, the main task here refers to the compensation of total expenses<br />
with the decision maker’s requirement regarding gross profits, without any<br />
change in the operating pattern of the clinic, as far as the manpower level is<br />
concerned.<br />
3.1. Specification of the Problem Decision Variables<br />
• The first 10 decision or choice variables X i (i = 1,...,10) are<br />
already defined in Table 3 and refer to the number of personnel in<br />
group i.<br />
— Variables X 11 ...X 20 , stand for the salary cost of each group.<br />
— Variable X 21 denotes total salary cost at the final year.<br />
— Variables X 22 ...X 31 refer to the total working hours, observed in<br />
the 5th column of Table 4.
Health Service Management 157<br />
• Variables X 32 ...X 35 refer to the average cost per patient, as it is stated<br />
in the first four rows of Table 2.<br />
— Variable X 36 denotes total per patient expenses.<br />
— Variables X 37 and X 38 refer to the equipment purchase and<br />
replacement.<br />
• Variable X 39 stands for the personnel insurance expenses.<br />
— Variable X 40 refers to the depreciation, laboratory, fixed and other<br />
expenses.<br />
• Variable X 41 is used to sum up all the above expenses (i.e., X 37 + X 38 +<br />
X 39 + X 40 )<br />
• X 42 , which also is a definition variable, is used to denote total operating<br />
cost for the final year t f (i.e., t 5 ) of the planning period.<br />
• Finally, variable X 43 denotes the average charge per patient, so that the<br />
desired gross profit would not be less than 750,000.<br />
3.2. Formulation of the Constraints (Goals)<br />
To facilitate the presentation, I classified the goals into the following seven<br />
categories.<br />
3.2.1. Personnel Employed<br />
The operation pattern of the clinic, regarding the personnel employed is<br />
presented in Table 1. Given that, the decision maker does not want to alter<br />
this pattern, I must first introduce the following 10 constraints (goals)<br />
X i + d − i<br />
− d + i<br />
= b i (i = 1,...,10) (5)<br />
where b 1 = 10, b 2 = 1, b 3 = 3, b 4 = 14, b 5 = 3, b 6 = 1, b 7 = 3, b 8 = 4,<br />
b 9 = 4 and b 10 = 10, are the desired targets, regarding the members of<br />
each group as shown in Tables 1, 3 and 4.<br />
It is useful to recall that di − and d<br />
i + denote the so-called deviational<br />
variables from the goals, which are complementary to each other, so that<br />
di − × d<br />
i + = 0. If the ith goal is not completely achieved, then the observed<br />
slack will be expressed by di − , which represents the negative deviation from<br />
the goal. On the other hand, if the ith goal is over achieved, then d<br />
i<br />
+ will<br />
take a non-zero value. Finally, if this particular goal is exactly achieved,<br />
then both di − and d<br />
i + will be zero.
158 Anna-Maria Mouza<br />
3.2.2. Personnel Claims for Salary Increase<br />
According to Eq. (2), I have to introduce the following 10 constraints in<br />
order to define the total salary expenses for each group.<br />
X 10+i − c i X i + d10+i − − d+ 10+i<br />
= 0, (i = 1,...,10). (6)<br />
It is recalled that coefficients c i are presented in Table 3. An additional<br />
variable X 21 is introduced to sum up salary expenses for all personnel at<br />
the final year. The computation of this total is based on Eq. (3).<br />
X 21 −<br />
20∑<br />
i=11<br />
X i + d21 − − d+ 21<br />
= 0. (7)<br />
3.2.3. Complete Utilization of Personnel Capacity in Terms<br />
of Working Hours<br />
The forecasted number of patients for the final year is not exceeding the<br />
clinic capabilities, regarding the existing infrastructure. Furthermore, the<br />
decision maker does not want to alter the existing operation pattern, since<br />
he considers that the present personnel manpower potential will be adequate<br />
to provide satisfactory service to the number of patients forecasted,<br />
provided that the under utilization of the regular working hours will be<br />
avoided. Hence, I introduced the following constraints in accordance to<br />
Eq. (4), which on the other hand may be regarded as a measure to provide<br />
job security to all personnel by avoiding underutilization of their regular<br />
working hours as shown in Table 4.<br />
X 21+i − w i X i + d21+i − − d+ 21+i<br />
= 0, (i = 1,...,10). (8)<br />
Note, the co-efficients w i , are also presented in Table 4.<br />
3.2.4. Patient expenses<br />
The axial-tomography expenses per patient can be expressed as:<br />
X 32 + d32 − − d+ 32 = 9.1.<br />
X-ray expenses per patient:<br />
X 33 + d33 − − d+ 33 = 5.7.<br />
Medical expenses per patient:<br />
X 34 + d34 − − d+ 34 = 59.4.<br />
(9a)<br />
(9b)<br />
(9c)
Administrative and miscellaneous expenses per patient:<br />
Health Service Management 159<br />
X 35 + d − 35 − d+ 35 = 32.1.<br />
Variable X 36 depicts total per patient expenses at the final year.<br />
X 36 − (X 32 + X 33 + X 34 + X 35 ) + d − 36 − d+ 36 = 0.<br />
(9d)<br />
(9e)<br />
3.2.5. Other Expected Expenses<br />
Installation of the IVR system.<br />
Computers and server replacement.<br />
X 37 + d − 37 − d+ 37 = 600.<br />
X 38 + d − 38 − d+ 38 = 400.<br />
(10a)<br />
(10b)<br />
The insurance funds.<br />
From Table 2, we see that the average expenses for personnel insurance<br />
amounts to 19% of the total employees’ salary. This is represented by the<br />
following constraint, taking into account the variable X 21 .<br />
X 39 − 0.19X 21 + d − 39 − d+ 39 = 0<br />
Depreciation, laboratory, fixed and other expenses<br />
X 40 + d40 − − d+ 40<br />
= (60,000 + 180,000 + 80,000). (10d)<br />
To sum all other expenses, I introduce the following relation.<br />
X 41 − (X 37 + X 38 + X 39 + X 40 ) + d − 41 − d+ 41 = 0.<br />
(10c)<br />
(10e)<br />
3.2.6. Total Expenses<br />
It is already mentioned that I introduce the variable X 43 in order to sum up<br />
total expenses, in the following way.<br />
X 42 − (X 21 + 3625X 36 + X 41 ) + d42 − − d+ 42<br />
= 0. (11)<br />
3.2.7. Total Gross Profit<br />
The gross profit achievement as set by the decision maker, may be introduced<br />
in the following way.<br />
3625X 43 − X 42 + d43 − − d+ 43<br />
= 750,000 (12)
160 Anna-Maria Mouza<br />
where X 43 , as it was mentioned earlier, represents the adequate charge per<br />
patient, in order to satisfy Eq. (12). It is recalled that the coefficient 3625,<br />
which appears in Eq. (11) too, is the stated forecast, regarding the expected<br />
number of patients for the final year t f .<br />
With the above formulation, we see that the number of constraints is<br />
equal to the number of choice variables (43).<br />
3.3. Priority of the goals — The objective function<br />
It is obvious that the decision maker, apart from determining the goals of<br />
the five-year plan, also has to specify the priorities (Gass, 1986), denoted by<br />
P i , in order to accomplish the optimum allocation of resources for achieving<br />
these goals in the best possible manner. One further point, that has to<br />
be considered in the formulation of the model, is the possibility of weighing<br />
the deviational variables at the same priority level (Gass, 1987). The<br />
list of decision maker’s goals is presented below in descending order of<br />
importance.<br />
1. To retain the present operating pattern of the clinic.<br />
This implies that no alterations are desired regarding the number and<br />
composition of the existing personnel (P 1 ). Different weights have been<br />
assigned to the various personnel groups.<br />
2. To achieve the desired level of gross profits. Also, to provide reserves for<br />
the personnel insurance, to cover all expenses for equipment replacements<br />
and installation of a new IVR system (P 2 ).<br />
3. To avoid underutilization of the personnel regular working hours, providing<br />
at the same time job security to all employees (P 3 ).<br />
4. To provide the funds necessary to cover all patient’s expenses (P 4 ).<br />
5. To provide adequate funds for covering laboratory and fixed expenses,<br />
depreciation etc. (P 5 ).<br />
6. To provide the wage increase, claimed by the personnel at the various<br />
groups (P 6 ).<br />
In this case, weights are used to discriminate the salary increase of<br />
each group. Heavier weights have been assigned to groups with lower<br />
salaries.<br />
According to the above ranking of the goals priority and taking into<br />
account the relations for the corresponding totals, the following objective
Health Service Management 161<br />
function is formulated.<br />
min z = (10P 1 d − 1 + 9P 1d − 2 + 8P 1d − 3 + 7P 1d − 4 + 6P 1d − 5 + 5P 1d − 6<br />
+ 4P 1 d − 7 + 3P 1d − 8 + 2P 1d − 9 + P 1d − 10 ) + (P 2d − 37 + P 2d − 38<br />
+ P 2 d − 39 + 3P 2d − 42 + P 2d − 43 ) + P 3<br />
31∑<br />
i=22<br />
di − ∑36<br />
+ P 4<br />
i=32<br />
+ (P 5 d − 40 + P 5d − 41 ) + (P 6d − 21 + 2P 6d − 11 + 2P 6d − 12 + 3P 6d − 13<br />
+ 4P 6 d − 14 + 5P 6d − 15 + 6P 6d − 16 + 7P 6d − 17 + 8P 6d − 18<br />
+ 9P 6 d19 − + 10P 6d20 − ). (13)<br />
d − i<br />
4. Initial Results<br />
With this specification of the goal programming model, I obtain the results<br />
as given on the left side of Table 5 (solution I). It is easily verified that all<br />
goals are achieved. Hence, this pattern dictated by the optimal solution is<br />
readily applicable. It should be pointed out that according to this solution,<br />
an average rate of total salary increase of about 4.9% is realized, whereas<br />
the mean salary per worker increases from 15,232.1 to 19,372.1. In fact, this<br />
average rate r (i.e., 4.9) has been computed from Eq. (1) in the following<br />
way, using natural logs.<br />
ln(x)<br />
5<br />
= ln(1 + r), where x = 1,026,720<br />
807,300<br />
and r = (y − 1) × 100, where y = e Z and Z = ln(x)<br />
5<br />
However, the charge per patient (∼740 ) is considered being higher,<br />
when compared to the average of the local market, affecting thus the<br />
competition level of the clinic.<br />
5. Imposition of a New Constraint<br />
Given the social character of health service, which has to be combined with<br />
the fact that the unit under consideration must be competitive, it is necessary<br />
to impose the following constraint.<br />
X 43 + d44 − − d+ 44<br />
= 700. (14)
162 Anna-Maria Mouza<br />
Thus, the total number of constrains has become 44. It should be noted<br />
that minimizing d<br />
44 + in the objective function, an upper limit of 700 is<br />
set to the average charge per patient. Hence, I added in the objective as in<br />
Eq. (13) the following term<br />
2P 2 d<br />
44<br />
+<br />
which implies that the restriction regarding the maximum average charge<br />
per patient has been considered as a second priority goal, properly weighed.<br />
5.1. Second-Run Results<br />
With the new constraint, the solution obtained is presented on the right<br />
side of Table 5 (solution II). I observe that constraint Eq. (14) is binding,<br />
and in order to achieve the level of gross profit set by the decision<br />
maker, the solution dictates that the total salary expenses must be reduced<br />
by 118,180.125, which is the value of the deviational variable d21 − , shown<br />
in Eq. (7). Evaluating the results of the second run, one may argue as to<br />
Table 5.<br />
Solution I (left) and solution II (right).<br />
No. Variable name Value No. Variable name Value<br />
The simplex solution (Number of patients = 3625)<br />
87 ×1 10.000 89 ×1 10.000<br />
88 ×2 1.000 90 ×2 1.000<br />
89 ×3 3.000 91 ×3 3.000<br />
90 ×4 14.000 92 ×4 14.000<br />
91 ×5 3.000 93 ×5 3.000<br />
92 ×6 1.000 94 ×6 1.000<br />
93 ×7 3.000 95 ×7 3.000<br />
94 ×8 4.000 96 ×8 4.000<br />
95 ×9 4.000 97 ×9 4.000<br />
96 ×10 10.000 98 ×10 10.000<br />
97 ×11 356,222.000 99 ×11 356,222.000<br />
98 ×12 33,054.000 100 ×12 33,054.000<br />
99 ×13 68,040.000 101 ×13 68,040.000<br />
100 ×14 199,290.000 102 ×14 199,290.000<br />
101 ×15 74,307.000 103 ×15 74,307.000<br />
102 ×16 36,190.000 104 ×16 36,190.000<br />
(Continued )
Table 5. (Continued )<br />
Health Service Management 163<br />
No. Variable name Value No. Variable name Value<br />
The simplex solution (Number of patients = 3625)<br />
103 ×17 39,476.000 105 ×17 39,476.000<br />
104 ×18 49,647.000 106 ×18 49,647.000<br />
105 ×19 50,643.000 107 ×19 50,643.000<br />
106 ×20 119,851.000 108 ×20 119,851.000<br />
107 ×21 1,026,720.000 21 d − 21<br />
118,180.125<br />
108 ×22 33,300.000 110 ×22 33,800.000<br />
109 ×23 2340.000 111 ×23 2340.000<br />
110 ×24 10,140.000 112 ×24 10,140.000<br />
111 ×25 29,120.000 113 ×25 29,120.000<br />
112 ×26 6240.000 114 ×26 6240.000<br />
113 ×27 2340.000 115 ×27 2340.000<br />
114 ×28 6240.000 116 ×28 6240.000<br />
115 ×29 8320.000 117 ×29 8320.000<br />
116 ×30 6240.000 118 ×30 6240.000<br />
117 ×31 15,600.000 119 ×31 15,600.000<br />
118 ×32 9.100 120 ×32 9.100<br />
119 ×33 5.700 121 ×33 5.700<br />
120 ×34 59.400 122 ×34 59.400<br />
121 ×35 32.100 123 ×35 32.100<br />
122 ×36 106.300 124 ×36 106.300<br />
123 ×37 600.000 125 ×37 600.000<br />
124 ×38 400.000 126 ×38 400.000<br />
125 ×39 195,076.797 127 ×39 172,622.578<br />
126 ×40 320,000.000 128 ×40 320,000.000<br />
127 ×41 516,076.812 129 ×41 493,622.562<br />
128 ×42 1,928,134.375 130 ×42 1,787,500.000<br />
129 ×43 738.796 131 ×43 700.000<br />
109 ×21 908,539.875<br />
whether this is an optimal solution, from the socio-economic point of view,<br />
since any reduction has to directly affect the employees salary solely. On<br />
the other hand, the decision maker is not willing to entirely undertake this<br />
reduction (118,180.125), decreasing gross profits by almost 16%. With all<br />
these in mind, I present the following proposition, which may be considered<br />
of particular importance.
164 Anna-Maria Mouza<br />
6. Some Alternatives<br />
After proper negotiations, the decision maker and the employees have to<br />
agree on the proportion, say a (where 0
Table 7.<br />
Salary at the initial and final years together with the annual rates of increase.<br />
Groups (i)<br />
Salary at the<br />
period t 0 ( )<br />
Initially<br />
claimed<br />
salary for the<br />
final year ( )<br />
Amount of<br />
reduction<br />
Salary to be<br />
realized at the<br />
final year ( )<br />
Actual annual<br />
rate of salary<br />
increase (%)<br />
New<br />
coefficients c i<br />
c i<br />
1 260,000 356,222 37,090.172 319,131.81 4.1836 31,913.181<br />
2 24,700 33,054 2859.187 30,194.81 4.0991 30,194.81<br />
3 54,600 68,040 2942.746 65,097.25 3.5795 21,699.083<br />
4 163,800 199,290 6384.697 192,905.3 3.3252 13,778.95<br />
5 58,500 74,307 4082.711 70,224.29 3.7209 23,408.096<br />
6 27,300 36,190 2689.870 33,500.13 4.1782 33,500.13<br />
7 32,760 39,476 961.173 38,514.83 3.2897 12,838.278<br />
8 41,600 49,647 858.898 48,788.10 3.2391 12,197.025<br />
9 42,640 50,643 567.861 50,075.14 3.2669 12,518.785<br />
10 101,400 119,851 652.748 119,198.25 3.2872 11,919.825<br />
Health Service Management 165
166 Anna-Maria Mouza<br />
increase for each group, the number of employees per group together with<br />
the ranking scores, I computed the index presented in the 4th column of<br />
Table 6. Multiplying the elements of this column by 59,090.0625, we get<br />
the entries of the 5th column.<br />
From the last column of Table 6, I computed the end period salary<br />
of each group and the corresponding annual rate of increase, which are<br />
presented in Table 7.<br />
From Table 7, we can see that the requirement for the annual rate of<br />
salary increase at each group to exceed 3.1%, is fully satisfied. In an opposite<br />
situation however, one has to reduce the value of “a” by, say, 0.05 and<br />
to repeat all calculations. It should be also noted that with the allocation<br />
proposed so far, total initial salary, spending will increase by an annual<br />
rate of 3.69% (from 807,300 to 967,629.9) and the annual wage per capita<br />
will increase by 19.86% (from 15,232.1 to 18,257.2). It is clear that all<br />
employees will be better off with smaller value of a. The composite index<br />
and all other entries of Tables 6 and 7 can be computed with the following<br />
code segment written in Visual Fortran.<br />
MODULE data_and_params<br />
! Goups=number of groups ! Planning_hor=Planning horizon (years)<br />
! R = Initial rate of salary increase at each group (vector)<br />
! Members = Number of employees at each group (vector)<br />
! Salary_in = Initial employees salary at each group (vector)<br />
! Salary_f = Final claimed salary for each group (vector)<br />
! Deficit = The value of deviational variable<br />
! a = Value of coefficient a (0
Health Service Management 167<br />
WRITE(out,*)’ Composite index and the final rate of salary increase’<br />
WRITE(out,’(/)’)<br />
Deficit=Total_Deficit*a ; sortR=R ; Salary=Salary_f/members<br />
! Perform ranking operation<br />
CALL SORTQQ(LOC(sortR),groups,SRT$REAL4)<br />
DO i=1,groups<br />
s=R(i)<br />
DO j=1,groups<br />
IF(sortR(j) == s) scores(i)= j<br />
ENDDO ; ENDDO<br />
! Compute composite index<br />
Rsmall=R/100. ; Y=Rsmall*Salary ; s=SUM(Y)<br />
Employees=SUM(Members)<br />
rate=(y*100)/s/100. ; X=rate*deficit/Employees<br />
s=SUM(X)<br />
rate=(x*100.)/s/100. ; s=SUM(X) ; X=rate*Members<br />
s=SUM(X)<br />
Rate=(X*100.)/s/100. ; s=SUM(Rate) ; X=rate*scores<br />
s=SUM(X)<br />
rate=(X*100)/s/100. ; s=SUM(rate) ; X=Deficit*rate<br />
! Compute revised annual rates of salary increase<br />
Y=Salary_f-X<br />
Logvector=(LOG(Y/Salary_in))/ planning_hor<br />
newrate=(EXP(logvector) -1.)*100.<br />
WRITE(out,*) ’ Index Reduction Final Salary New rate’<br />
DO i=1,groups<br />
WRITE(out,’(1X,F9.6,2X,F10.3,4X,F10.2,6X,F6.4)’) Rate(i)&<br />
&, X(i), Y(i), newrate(i)<br />
ENDDO<br />
END<br />
7. Final Results<br />
In the last column of Table 7, the revised coefficients c i are stated. These<br />
new coefficients are to be considered in Eq. (2), and in the corresponding<br />
restrictions seeing in Eq. (6) for the final run. Also, the objective in Eq. (13)<br />
has been slightly reformed as follows:<br />
min z = (10P 1 d1 − + 9P 1d2 − + 8P 1d3 − + 7P 1d4 − + 6P 1d − 5 + 5P 1d6<br />
−<br />
+ 4P 1 d − 7 + 3P 1d − 8 + 2P 1d − 9 + P 1d − 10 ) + (P 2d − 37 + P 2d − 38<br />
+ P 2 d − 39 + 3P 2d − 42 + 2P 2d + 44 ) + P 3<br />
31∑<br />
i=22<br />
di − ∑36<br />
+ P 4<br />
+ (P 5 d − 40 + P 5d − 41 ) + (P 6d − 21 + 2P 6d − 11 + 2P 6d − 12<br />
i=32<br />
+ 3P 6 d − 13 + 4P 6d − 14 + 5P 6d − 15 + 6P 6d − 16 + 7P 6d − 17 + 8P 6d − 18<br />
+ 9P 6 d − 19 + 10P 6d − 20 ) + P 7d − 43 .<br />
d − i
168 Anna-Maria Mouza<br />
The solution obtained with the re-stated restrictions satisfies all requirements.<br />
Now, total expenses sum up to 1,857,817.125. It has to be underlined,<br />
however, that since the value of the deviational variable d43 − is<br />
70,317.2, it is implied that total gross profits undergo a reduction of about<br />
9.38% (from 750,000 to 679,682.8), which is a bit higher when compared<br />
to the corresponding percentage (∼6%) of total salary reduction. Nevertheless,<br />
the results provide a sound basis to achieve an optimal solution, in the<br />
sense that it is acceptable by all people involved without any violations of<br />
the existing rules.<br />
8. Conclusions<br />
Many clinics like the one presented here benefit the patients, the health<br />
service in a convenient way, and help attain consistency and continuity<br />
of treatment. These achievements are heavily dependent on a successful<br />
budget planning. The characteristics of a model of this type for a clinic<br />
are based upon many factors, such as the type of the clinic, the location,<br />
the size, the medical specialty, etc. Hence, it is difficult to design a general<br />
model that can be applied to all types of clinics. However, once a budget<br />
planning model has been developed, it can be easily modified at a later<br />
stage to fit many other types of clinics. With this in mind, I consider in this<br />
paper an orthopedic clinic and implemented a model designed for a fiveyear<br />
planning period, using the goal programming approach of aggregate<br />
budget planning for this particular health care unit. To start with, the goals<br />
and the relevant priorities are set in advance, to formulate the corresponding<br />
goal programming model. The solution obtained is operational in the sense<br />
that, through the analysis of resource requirements, the acceptable charge<br />
per patient, the desired level of gross profits, the claimed rate of salary<br />
increases by the employees, and the trade-offs among the set of goals are<br />
achieved with a minimum sacrifice. The model shows that this sacrifice is<br />
negotiable between different sides, so that the business manager can finally<br />
establish an aggregative budget planning with the best perspectives, since<br />
the realization of a fair wage policy can be obtained through using the<br />
composite index, introduced in this paper.<br />
References<br />
Charnes, A and W Cooper (1961). Management <strong>Models</strong> and Industrial Applications of<br />
Linear Programming (Vols. 1 and 2). New York: Wiley.<br />
Everett, JE (2002). A decision support simulation model for the management of an elective<br />
surgery waiting system. Health Care Management Science 5, 89–95.
Health Service Management 169<br />
Flessa, S (2000). Where efficiency saves lives: a linear programme for the optimal allocation<br />
of health care resources in developing countries. Health Care Management Science 3,<br />
249–267.<br />
Gass, SI (1986). A process for determining priorities and weights for large scale linear goal<br />
programming. Journal of Operational Research Society 37, 779–785.<br />
Gass, SI (1987). The setting of weights in linear goal programming. Computers and Operations<br />
Research 14, 227–229.<br />
Mouza and Anna-Maria (1996). An economic approach of the Greek health sector, based<br />
on domestic data. Modeling the hospital operation process. Ph.D. Theses, Greece:<br />
Aristotle University of Thessaloniki.<br />
Mouza and Anna-Maria (2002). Estimation of the total number of hospital admissions and<br />
bed requirements for 2011: the case for Greece. Health Service Management Research<br />
15, 186–192.<br />
Mouza and Anna-Maria (2003). IVR and administrative operations in healthcare and hospitals.<br />
Journal of Healthcare Information Management 17(1), 68–71.<br />
Mouza and Anna-Maria (2006). A note regarding the projections of some major health<br />
indicators. Quality Management in Health Care 15(3), 184–199.<br />
Ogulata, SN and R Erol (2003). A hierarchical multiple criteria mathematical programming<br />
approach for scheduling surgery operations in large hospitals. Journal of Medical<br />
Systems 27(3), 259–270.
This page intentionally left blank
Chapter 5<br />
Modeling National Economies
This page intentionally left blank
Topic 1<br />
Inflation Control in Central and Eastern<br />
European Countries<br />
Fabrizio Iacone and Renzo Orsi<br />
University of Bologna, Italy<br />
1. Introduction<br />
The recent accession to the European Union (EU) of 10 new members<br />
officially opened the agenda of their participation in the European Monetary<br />
Union (EMU) too.<br />
Although, the accession of the new European partners to the euro-area<br />
is formally analyzed, keeping the Maastricht criteria as the main reference,<br />
these do not necessarily reflect the effective integration and convergence of<br />
the economies. With this respect, we think it is important to study if and how<br />
the Eurosystem can successfully control inflation after the extension of the<br />
EMU: we then investigated the monetary policy transmission mechanism in<br />
the Central and Eastern European Countries (CEECs) to see how compatible<br />
it is with one of the current members of euro-area.<br />
Knowledge of the mechanism of transmission of monetary policy is<br />
also necessary to the CEECs to choose the monetary strategy for the period<br />
preceding the accession to the euro-area. The discussion is open, both in<br />
the literature and among monetary institutions, on whether a direct inflation<br />
targeting or some kind of exchange rate commitment should be preferred:<br />
by comparing the mechanism of transmission of policy impulses in different<br />
regimes, we can see if a certain strategy made inflation control easier or<br />
more difficult, and assess its cost by looking at the impact on the economic<br />
activity.<br />
We focused on Poland, Czech Republic, and Slovenia because they<br />
are very different for the initial conditions from which they started the<br />
transition to functioning market economies and later the integration in the<br />
EU, for the relative size and the degree of openness to the international<br />
trade, for the policies implemented in due course and even for their historical<br />
173
174 Fabrizio Iacone and Renzo Orsi<br />
developments: we think that because of these differences they are, together,<br />
representative of the whole group of CEECs. For the euro-area, we took<br />
Germany as the reference country.<br />
2. Inflation Control over 1989–2004<br />
2.1. Transition, Integration, and Accession<br />
Although Poland, Czech Republic, and Slovenia shared the goals of implementing<br />
a functioning market economy and of acceding to the EU, the<br />
policies they implemented to reach them were rather different.<br />
The first reason for the difference is in the productive structure at the<br />
beginning of the transition. Yugoslavia established a quasi-market system<br />
well before 1989, with quite independent firms and relatively hard budget<br />
constraints; production was much more centralized and budget constraints<br />
much softer in Poland and Czech Republic, and, to make things tougher for<br />
these two countries, they also had to suffer the break-up of the council for<br />
mutual economic assistance (CMEA), in which they were well integrated<br />
(Czech Republic also had to endure the additional trade disruption, caused<br />
by the separation from Slovakia). Poland and Czech Republic represent<br />
an average position with respect to the general condition of the productive<br />
structure of the CEECs at the beginning of the transition. Local differences<br />
remain: for example, Hungary implemented some timid reforms before<br />
1989, while the three Baltic countries experienced a much bigger shock,<br />
because the disintegration of the Soviet Union resulted in the loss of their<br />
largest trade partner. The stronger centralization, the softer budget constraint,<br />
and the higher disruption caused by the break. up of the CMEA,<br />
and of Czechoslovakia could have coeteris paribus caused a more turbulent<br />
and longer transition for Poland and Czech Republic, but in reality Slovenia<br />
had to face the major shock due to the collapse of Yugoslavia and of the<br />
subsequent wars, which deprived the country of its biggest market and had<br />
other adverse effects including the loss of revenues and hard currency due<br />
to the fall of tourism.<br />
A more precise ranking emerges when the size and the openness to<br />
international trade is considered, with Poland being the largest country and<br />
Slovenia the smallest one. As far as the monetary policies implemented for<br />
inflation stabilization, all the CEECs except Slovenia, began their transition<br />
with a strong commitment to a fixed exchange rate. The initial price<br />
liberalization, in fact, left the system without a nominal anchor, and, also
Inflation Control in Central and Eastern European Countries 175<br />
because of a natural price rigidity with respect to negative corrections, the<br />
re-alignment of relative prices took the form of a sudden increase and a steep<br />
inflation followed: by opening to the international trade (and by introducing<br />
the proper legislation, in order to force the local producers to face a harder<br />
budget constraint), the countries in transition had a chance to import a price<br />
structure similar to the one of their commercial partners. In fact, being the<br />
exchange rate fixed, the domestic producers could not raise their prices too<br />
much without losing competitiveness with respect to the foreign ones.<br />
Slovenia followed a different approach to disinflation, targeting the<br />
growth of a relatively narrow monetary aggregate (M1); according to Capriolo<br />
and Lavrac (2003), anyway the key characteristic of the period was the<br />
continuous intervention on the foreign exchange market in order to prevent<br />
an excessive real rate appreciation (contrary to the strategy of the Bundesbank<br />
and of the Eurosystem, the control of M1 was also enforced with nonmarket<br />
instruments and procedures). Since mainstream optimal currency<br />
area literature prescribes a greater incentive to adopt a stable exchange rate<br />
to the countries more open to the international trade, the fact that the outcome<br />
is actually reversed means that price stabilization rather than trade<br />
was the priority in the exchange rate commitment served in Poland and in<br />
Czechoslovakia.<br />
Fixing the exchange rate may have contributed to price stabilization,<br />
and inflation actually dropped very quickly, but there are certain differences<br />
remained with respect to the EU, leading to a strong real exchange rate<br />
appreciation. This may have been partially due to the Balassa-Samuelson<br />
effect: assuming that the marginal productivity and the wage in the sector<br />
of internationally traded goods, is set on the foreign market and that the<br />
same wage is transferred to the production of the non-traded goods, then<br />
the productivity gap between the two sectors is compensated by a price<br />
increase in the less productive domestic sector. This yields a progressive<br />
appreciation of the real exchange rate and higher domestic inflation.<br />
The faster growth of productivity, determined by the transfer of<br />
technology from the international trade partners and the more efficient allocation<br />
of the resources within the economies, compensated part of the pressure<br />
on the domestic producers, but the fixed exchange rate arrangements<br />
were not sustainable for a long time: Poland switched to a regime of preannounced<br />
crawling peg as early as 1991, later coupling it with an oscillation<br />
band which was gradually widened until it was finally abandoned in 2000;<br />
Czech Republic resorted to a free float without explicit commitment as<br />
early as 1997 under the pressure of a speculative attack, its currency having<br />
progressively over-evaluated in real terms over time. Both the countries
176 Fabrizio Iacone and Renzo Orsi<br />
replaced the monetary anchor by introducing a direct inflation target (Czech<br />
Republic starting in 1998, Poland in 1999). The progressive weakening of<br />
the exchange rate commitment took place in Hungary and Slovakia too, but<br />
in these last two countries, the switch of targets was not complete: Hungary<br />
retained a fixed exchange rate commitment, albeit with a very wide band,<br />
while Slovakia did not implement an explicit inflation targeting because of<br />
the high role that administrated prices still had in the consumer prices. Anyway<br />
this is not an unanimous attitude: the three Baltic States kept a fixed<br />
exchange rate with an extremely tight band, apparently trying to exploit the<br />
credibility acquired with it during the first phase of the transition, and Estonia<br />
even withstood a moderate speculative attack rather than devaluating.<br />
Once again, Slovenia went in the opposite direction: according to Capriolo<br />
and Lavraac, a more aggressive exchange rate strategy has been pursued<br />
since 2001, with an active, albeit informal, crawling peg.<br />
2.2. Exchange Rate and Direct Inflation Targeting<br />
The fixed exchange rate played a key role in the stabilization phase; nowadays,<br />
it is still the reference of monetary policy in the Baltic States, and it is<br />
also present in Hungary. Since, the exchange rate commitment is also part of<br />
the Maastricht criteria, where two years of participation to the exchange rate<br />
mechanism (ERM) 2 is required, the decision of Poland, Czech Republic,<br />
and Slovakia, to completely abandon it may indeed seem to be a paradox.<br />
Surely, a small country cannot ignore the effect of the exchange rate<br />
fluctuations on the stability of financial institutions and the economy as<br />
a whole, but a commitment to a fixed parity is a much stronger policy.<br />
Whether, macroeconomic stability and inflation control are helped or<br />
hindered by this depends on the credibility of the commitment itself.<br />
There can be little doubt that an exchange rate commitment is a risky<br />
policy: the integration of the financial markets allows the mobilization of<br />
large amounts of funds, and no central bank can muster enough foreign<br />
exchange reserves to resist a sustained attack, nor indeed may desire to do<br />
it, because the mere defence of the currency could require a substantial rise<br />
in the domestic interest rates, for which the economy may be even more<br />
detrimental than the devaluation that it was trying to prevent. Against an<br />
exchange rate commitment, stands the evidence of speculative attacks in<br />
the past: consider, for example, the break up of the ERM 1, when even<br />
the French currency suffered some pressure, although the fiscal, financial,<br />
and macroeconomic situation of the country was not worse than it was<br />
in Germany. The same lesson can indeed be derived by looking at the
Inflation Control in Central and Eastern European Countries 177<br />
Hungarian experience: when the two objectives were in conflict, as in June<br />
2003, one had to be abandoned, and a devaluation took place.<br />
Since the accession to the EU also required the complete adoption of<br />
the acquis communautaire, thereby including the elimination of capital<br />
controls, the apparent paradox of several CEECs switching to a free float<br />
and inflation targeting, for the period preceding the adoption of the euro,<br />
is then a strategy that may protect their currencies from unrealistic parities<br />
and speculative attacks.<br />
On the other hand, a direct inflation targeting can only be successful if<br />
the monetary authority has earned itself a solid reputation, and this primarily<br />
requires the independence from any political interference. Since the CEECs<br />
only acquired a formal central bank independence in the last few years, as a<br />
part of the process of adoption of the acquis communautaire, their credibility<br />
may still be weak. Central bank independence is even more a concern,<br />
when the definition is extended beyond the mere legal requirement: in most<br />
of the cases for example, the credibility was seriously weakened because<br />
the fiscal authority had the opportunity to partially finance its expenditures<br />
through the central bank; political pressures, as experienced in the Czech<br />
Republic and in Poland, may undermine the credibility of the central bank<br />
even when they are resisted, because they may erode the consensus in<br />
the country. Several indices of central bank independence, were proposed<br />
in the literature: according to Maliszewski (2000), and to the extensive<br />
survey in Dvorsky (2000), the CEECs were lagging behind with respect<br />
to their Western counterparts, Poland and the Czech Republic faring better<br />
than Slovenia. Indeed, it is exactly because the credibility of the monetary<br />
authorities is lower in the CEECs that Amato and Gerlach (2002) proposed<br />
to supplement the inflation targeting with a mild monetary commitment,<br />
this would increase the information in the market.<br />
It is then important to ascertain if the exchange rate commitment is a<br />
risk worth taking, not only for the countries on the way to EMU that are still<br />
more than two years away from the formal discussion of their convergence<br />
according to the Maastricht criteria, but also for other emerging economies<br />
in the world.<br />
2.3. A Summary of the Results of Previous Analyses<br />
Several empirical analyses on inflation control in CEECs appeared in the<br />
recent years. Iacone and Orsi (2004) surveyed many applied works: these<br />
varied for the country that was analyzed, for the period considered and for<br />
the methodology adopted, making comparison rather difficult.
178 Fabrizio Iacone and Renzo Orsi<br />
Interpretation of the results was often dubious because either the variables<br />
of interest were replaced by first or second differences, or the intermediate<br />
steps linking the monetary instrument to the target were omitted,<br />
thus obscuring any potential evidence about the channel through which<br />
inflation may be controlled. Reliability was also hampered by the fact that<br />
in many cases no stability analysis was provided, despite the potentially<br />
disruptive effects of the transition on the estimates, nor any attempt was<br />
made to model the slow formation of a market economy despite this being<br />
the main feature of the period.<br />
It seems nevertheless fair to conclude that the exchange rate is very<br />
important for the stabilization of inflation, an appreciation of the real<br />
exchange rate reducing the growth rate of prices. The empirical evidence<br />
supporting the existence of a standard interest rate channel was much more<br />
controversial, the results depending largely on the econometric approach<br />
and model adopted by the researcher and on the sampling period considered.<br />
2.4. A Structural Model for Monetary Policy<br />
It is possible that the weak support provided by these analyses is at least<br />
partially due to the methodology adopted: VAR models have the advantage<br />
of requiring a minimal structural specification, but, in return, they need<br />
the estimation of many parameters. Considering the short length of the<br />
sampling size and the potential extreme instability of the parameters due to<br />
the transition, large standard errors and non-significant estimates seem to<br />
be unavoidable consequences.<br />
A structural model can provide a more parsimonious approach: we<br />
followed Rude-bush and Svensson (1999) and considered the model<br />
⎧<br />
⎨ π t = α 0 + α π1 π t−1 + α π2 π t−2 + α π3 π t−3<br />
+ α π4 π t−4 + α y y t−1 + ε π,t PC<br />
⎩<br />
y t = β 0 + β y y t−1 − β<br />
AD<br />
where y t is a measure of the economic activity, π t is the annualized inflation<br />
within the quarter, it is a short-term interest rate, ¯π =<br />
4 1 ∑ 3j=0<br />
π t−j is the<br />
average inflation rate in the last four quarters (i.e., the inflation over the<br />
last year), and ī =<br />
4 1 ∑ 3j=0<br />
i t−j is the average interest rate over the same<br />
period.<br />
The equations are referred to as Phillips Curve (PC) and aggregate<br />
demand (AD), respectively because the first one can be given a structural<br />
interpretation as a PC while the second one may describe the transmission<br />
of monetary policy on the economic activity quite like in an AD function.
Inflation Control in Central and Eastern European Countries 179<br />
With this specification, agents extrapolate from the past inflation to formulate<br />
the current period price level. The mechanism of transmission of<br />
monetary policy to inflation is indirect: an expansionary monetary policy<br />
takes place as a decrease in the real rate ī t−1 −¯π t−1 , which boosts the<br />
economic activity in the AD equation; in the second moment, the pressure<br />
on the demand enhances inflation, as illustrated in the PC equation.<br />
The instrument and the mechanism of transmission of monetary policy are<br />
then consistent with the operating procedures of both the FED and of the<br />
Bundesbank, and later of the ECB.<br />
Rudebush and Svensson (1999) allowed for an additional term β y2 y t−2<br />
in the AD equation, but we found that, possibly also due to the smaller<br />
dimension of our samples and the high collinearity with β y2 y t−2 , the contribution<br />
of the two different effects in general could not be distinguished.<br />
We, then considered the proposed AD equation sufficient, also because<br />
one lag proved to be enough to have no serial correlation in the residuals.<br />
Rudebush and Svensson also specified the model without intercepts α 0 ,β 0 ,<br />
having formulated for mean-corrected data, but we preferred to allow for<br />
the constant and test, for its possible exclusion explicitly.<br />
Rudebush and Svensson (2002) remarked that ∑ 4<br />
j=1 α πj = 1 can be<br />
expected as it corresponds to a vertical PC in the long run. They also<br />
explicitly discussed the fact that the backward looking formulation of<br />
the PC and AD equations imply adaptive rather than rationale expectations,<br />
but they argued that this may well be the case especially in a phase<br />
of monetary and inflationary instability; as an alternative, they suggested<br />
that the given specification may simply be considered as reduced form<br />
equations.<br />
Notice that even if the constraint ∑ 4<br />
j=1 α πj = 1 is applied, the model<br />
still does not necessarily impose a unit root to inflation, because the dynamics<br />
of y t has to be taken into account too: consider for example α π1 = 1,<br />
α π2 = α π3 = α π4 = 0, β y = 0, and the monetary policy rule ī t =¯π t + π t ;<br />
the PC equation then becomes β r ∈ y, t+ ∈π, t, thus, inducing a stationary<br />
behavior. It is indeed, precisely by steering the output gap using the interest<br />
rate that the stabilization of inflation is ultimately possible.<br />
This model may well be applied to the transition economies too, especially<br />
considering that the accession of these countries to the EU requires<br />
them to adopt a monetary policy setting that is institutionally compatible<br />
with the one of the ECB. In the case of a small open economy, though,<br />
two other factors should be taken into account: the demand of domestic<br />
products generated by the trade partners and the effect of the international<br />
competition on local prices.
180 Fabrizio Iacone and Renzo Orsi<br />
Following Svensson (2000), we then augmented the model by introducing<br />
the level of the economic activity in the international partners wt and the<br />
percent real exchange rate depreciation (q t −q t−1 ), where q t = p ∗ t +e t −p t<br />
is the logarithm of the real exchange rate and p t ,p ∗ t and e t are the logarithm<br />
of the local and foreign level of prices and of the nominal exchange<br />
rate, respectively. Both the variables are added to the AD equation, and<br />
the real exchange rate is added to the PC equation as well. For the AD,<br />
the assumption is that phases of high economic activity abroad result in<br />
higher demand of locally produced goods, while a too high level of prices<br />
(in real terms) causes a shift of the domestic and foreign demand towards<br />
the external producers. The real exchange rate entered the PC equation,<br />
because the international competition imposes some price discipline to the<br />
local producers: for given foreign prices, a nominal exchange rate appreciation<br />
makes foreign goods cheaper in local currency, thus, forcing the<br />
domestic producers to lower their prices to keep up with the external competitors,<br />
while for given nominal exchange rate higher foreign prices give<br />
the domestic producers to set higher prices and stay in the market; we also<br />
refer to Svensson (2000), where he explains how an increase in foreign<br />
prices in local currency, either due to the move of the foreign price itself or<br />
to the exchange rate, results in higher cost of inputs and then feeds back in<br />
higher prices of output.<br />
Maintaining the autoregressive structure of Rudebush and Svensson,<br />
we describe the economy with the two equations model<br />
π t = α 0 + α π1 π t−1 + α π2 π t−2 + α π3 π t−3<br />
+ α π4 π t−4 + α y y t−1 − α q (q t−1 − q t−5 ) + ε π,t PC<br />
y t = β 0 + β y y t−1 − β w w t−1 − β q (q t−1 − q t−5 ) + ε y,t AD<br />
where we actually considered relevant for the exchange rate the depreciation<br />
over the whole year.<br />
In the original model, Svensson considered the real exchange rate rather<br />
than the percent depreciation: in our model specification, we preferred the<br />
latter because as we saw the real exchange rates of most of the CEECs<br />
appreciated since the beginning of the transition without resulting in a<br />
dramatic decline of the economic activity.<br />
Moreover, we kept the AD/PC structure as a general reference, but<br />
given that the specification is somewhat arbitrary, we tried to adapt it to<br />
the economies considered. Our sample starts from 1990, and to avoid the<br />
risk of model instability in the very first part of the sample, the data were<br />
used only to compute the output gap or as lags. In some cases, we did
Inflation Control in Central and Eastern European Countries 181<br />
not find adequate data so the sample period was even shorter. Data are<br />
monthly, but in order to reduce the volatility, we used quarterly averages,<br />
adjusting the frequency properly. We took the CPI to compute inflation and<br />
the real exchange rate, and the seasonally adjusted industrial production<br />
to derive a measure of economic activity (when only the nonseasonally<br />
adjusted series was available, we run a preliminary step using the X-12<br />
filter in EViews to remove seasonality); for the nominal interest rate we<br />
used a three-months inter-bank rate or the Treasury bills rate of the same<br />
maturity. More details on the data used are at the end of this paper. We<br />
always considered Germany as the international reference, even if in the<br />
first part of the sample period some countries pegged the exchange rate to<br />
the US$: the EU in fact, constitutes by far the biggest trade partner for the<br />
CEECs.<br />
There is a strong seasonal component in the quarterly inflation, which<br />
is likely to shift the weights towards the fourth lag in the PC equation;<br />
the choice of the year appreciation of the real exchange rate was then also<br />
convenient to remove the seasonal effect.<br />
As a measure of the economic activity we took the output gap, defined<br />
the difference between the observed and the potential output computed<br />
using the Hodrik Prescott (HP) filter on the quarterly average of the production<br />
index. Even if the HP gap is usually considered an inexact measure<br />
of the economic cycle, due to its purely numerical definition, we still think<br />
that it is at least informative of the effective output gap. Plots of the inflation<br />
in the four countries are in Fig. 1 (figures are all collected at the end of<br />
Figure 1.<br />
Yearly inflation, Germany, Poland, Czech Republic, and Slovenia.
182 Fabrizio Iacone and Renzo Orsi<br />
the work). The inflation used for this graph was defined as the growth rate<br />
of prices on the previous year: we presented it albeit, we used a different<br />
measure in our empirical study because it is very common in the descriptive<br />
analysis in the literature. Clearly, though, averaging the inflation in this way<br />
induces spurious autocorrelation in the data, so in the empirical analysis we<br />
measure inflation as the annualized growth rate of prices over each period.<br />
We plotted this and the output gap for Germany in Fig. 2; in Figs. 3–5,<br />
we plot the same data and the yearly real exchange rate depreciation for<br />
Poland, Czech Republic, and Slovenia, respectively. From these figures, it<br />
Figure 2.<br />
Quarterly inflation and output gap, Germany.<br />
Figure 3.<br />
Quarterly inflation, output gap, real exchange rate depreciation, Poland.
Inflation Control in Central and Eastern European Countries 183<br />
Figure 4.<br />
Republic.<br />
Quarterly inflation, output gap, real exchange rate depreciation, Czech<br />
Figure 5.<br />
Slovenia.<br />
Quarterly inflation, output gap, real exchange rate depreciation,<br />
is already possible to notice that, while for Germany, the dynamics of inflation<br />
is dictated by the output gap, in the three new EU members inflation<br />
is largely dictated by the real exchange rate dynamics.<br />
Estimation was carried out using OLS equation by equation, as in Rudebush<br />
and Svensson (1999). We did not introduce any other modification to
184 Fabrizio Iacone and Renzo Orsi<br />
the AD equation, but we re-wrote the PC as<br />
π t = α 0 + ρ 0 π t−1 + ρ 1 π t−1 + ρ 2 π t−2 + ρ 3 π t−3 + ρ y y t−1<br />
+ α q (q t−1 − q t−5 ) + ε π,t PC<br />
(the term q t−1 − q t−5 does not appear in the equation for Germany). This<br />
is a re-parametrization of the original equation, with<br />
ρ 0 = (α π1 + α π2 + α π3 + α π4 − 1)<br />
ρ 1 =−(α π2 + α π3 + α π4 ), ρ 2 =−(α π3 + α π4 ), ρ 3 =−α π4<br />
Being simply a re-parametrization, the residual sum of squares that<br />
has to be minimized is the same, so the OLS estimates are not different<br />
than the estimates that one would obtain in the original model in levels.<br />
The advantage of this specification is that long and short dynamics are<br />
explicitly separated, ∑ 4<br />
j=1 α πj = 1 corresponding to p 0 = 0, and that,<br />
more importantly, the multi-collinearity due to the terms π t−1 to π t−4 is<br />
much less severe.<br />
Correct specification was primarily checked by looking at the tests for<br />
normality (Jarque and Bera), autocorrelation (up to the fourth lag: Breusch<br />
.Godfrey LM) and heteroskedasticity (white without cross terms) in the<br />
residuals: we referred to these tests using N for the normality, AC for the<br />
autocorrelation, and H for the heteroskedasticity. If the model appeared to<br />
be correctly specified, we then analyzed the stability of the parameters over<br />
time. If we did not reject the hypothesis of stability either, we moved to<br />
test specific restrictions on the parameters, and to estimate a more efficient<br />
version of the model. In all the tests, we used the conventional 5% size. In<br />
the tables in the Appendix, we presented both the P-values associated to<br />
the tests and a summary of the equation estimated when the restriction was<br />
imposed.<br />
We considered the normality test first, because we used its result to<br />
choose between the small or large sample version of the other tests. Since<br />
the small and large sample statistics are asymptotically equivalent in large<br />
samples, the asymptotic interpretation of the results is not affected. Admittedly,<br />
due to the asymptotic nature of the Jarque and Bera test, and to the<br />
small sample lower order bias in the estimation of autoregressive coefficients,<br />
the reader may be cautious in the interpretation of the results, especially,<br />
when only portions of the dataset are used to estimate parameters<br />
(like in the Chow stability test or in the estimates on sub-samples): notice<br />
anyway that this is a problem that depends on the dimension of the sample;<br />
so, indeed the same caveat applies to all the empirical analyzes present in
Inflation Control in Central and Eastern European Countries 185<br />
the literature. The reader may then want to consider the estimates and the<br />
statistics computed on the estimated parameters more as general indications<br />
in small samples (especially, if the sub-sample 1998–2004 only is used),<br />
but we think that despite these drawbacks, the estimates and the tests do<br />
still provide some information on the structure that we intend to analyze,<br />
albeit an approximate one.<br />
Autocorrelation in the residuals is a serious mis-specification since it<br />
results in inconsistent estimates given that both the equations include a<br />
lagged dependent variable. Heteroskedasticity is a minor problem, because<br />
estimation is simply inefficient (compared to the GLS estimator) but not<br />
inconsistent. Yet, since the standard errors indicated by the regression are<br />
not correct, evidence of heteroskedasticity requires at least a different estimation<br />
of the variance-co-variance matrix of the estimates.<br />
Due to the peculiarity of the transition, we think it is particularly important<br />
to control the parameter stability over time, and possibly to model it<br />
when apparent. We used as preliminary diagnostic the CUSUM squared<br />
test, but also considered as a particular breakpoint in 1998, because of the<br />
opening of the negotiation to access the EU. Clearly, this is just a broad<br />
indication: one could also argue that by that time the negotiation opened<br />
the transition to a market economy was already nearly complete, and the<br />
issue was convergence to the EU standards. Yet, that period may still be<br />
informative about a potential break even if it was not located exactly there.<br />
We also considered country-specific breakpoints in the PC equations to<br />
analyze the effect on the inflation dynamics of monetary regime changes:<br />
for Germany, we allowed a break in 1999Q1, when the control of monetary<br />
policy was transferred from the Bundesbank to the ECB, because it may<br />
have been perceived as a weakening of the anti-inflationary stance, inducing<br />
higher inflationary expectations; for Poland, we considered a break in<br />
2000Q2, when the exchange rate commitment was formally abandoned;<br />
for Slovenia, the change is in 2001Q1, when the approach towards the real<br />
exchange rate became effective. We tested the presence of the break with a<br />
standard Chow test, unless the sample after the break was very little, in this<br />
case we used the Forecast test, which is exactly designed to test for breaks<br />
when only a little data are located after the break.<br />
If the model was correctly specified and stable, we improved the efficiency<br />
of the estimates by removing the variables whose coefficients were<br />
estimated with a sign opposite to the one prescribed in the economic theory<br />
(we assumed that α y , α q , β y , β w , β q , β r are not negative), and by imposing<br />
other restrictions (including ∑ 4<br />
j=1 α πj = 1): we tested these with a Wald<br />
statistic, and referred to these tests as W in the tables.
186 Fabrizio Iacone and Renzo Orsi<br />
Of course, given the small dimension of the samples and the potential<br />
instability of the models due to the transition, a certain caution should still<br />
be exerted in the interpretation of the results. But, we think that even in<br />
this case our analysis is at least able to get a general, broad picture of the<br />
situation: the lack of precision that may be attributed to the small dimension<br />
of the sample, for example, is already taken into account in the standard<br />
error. If, under normality, a test statistic rejects the null hypothesis despite<br />
the large variability due to the small sample, then we would still regard the<br />
estimate as informative, although may be only on the sign.<br />
3. The Empirical Evidence<br />
3.1. Germany<br />
As a benchmark case, we first estimated the model for Germany over the<br />
period 1991Q1–2004Q1. The choice of the sample is two-fold: first, only<br />
data after the German re-unification were considered, leaving many of the<br />
problems related to that particular transition out of the discussion; second,<br />
this time span is roughly the same one covered by the data of the CEECs in<br />
most of the applied analyzes, making the results comparable in this sense.<br />
We called this a “benchmark” model because it is widely assumed that this<br />
transmission mechanism for monetary policy was at work in Germany in<br />
the years we are interested in. Furthermore, Germany is geographically<br />
and economically close to the CEECs, accounting for a large share of their<br />
international trade. Finally, it is an interesting reference because Germany<br />
is the biggest country in the euro-area: if the monetary transmission in the<br />
CEECs works in a very different way, their integration in the euro may<br />
adversely affect the dynamics of inflation.<br />
The estimated model for Germany was<br />
⎧<br />
ŷ ⎪⎨ t = 0.011 + 0.769y t−1<br />
(0.005) (0.084)<br />
⎪⎩ ̂π t =−1.02π t−1<br />
(0.10)<br />
− 0.0038y t−1<br />
(0.00163)<br />
− 0.86π t−2<br />
(0.12)<br />
− 0.71π t−3 + 17.62y t−2<br />
(0.10) (12.18)<br />
over the period 1991Q2–2004Q1 for the AD, and 1993Q1–2004Q1 for<br />
the PC.<br />
According to the diagnostic tests, the AD equation was correctly specified<br />
and stable over the whole sample; the CUSUMsq hit the boundary in<br />
the period 1993–1996, but we found no evidence of break with the Chow
Inflation Control in Central and Eastern European Countries 187<br />
test (further details about all the tests and a summary of the estimates are<br />
in the tables at the end).<br />
The inflation dynamics, on the other hand, was not stable when the<br />
whole sample was taken into account. The instability, though was clearly<br />
restricted to the first part of the sample, and could indeed be removed dropping<br />
the years 1991 and 1992. It is interesting to observe that it is exactly<br />
the period of the recovery of the inflationary shock due to the German<br />
re-unification, a phenomenon that was usually regarded as temporary and<br />
exceptional. Diagnostic tests confirmed that the equation estimated using<br />
data from 1993 onwards was correctly specified and stable; notice anyway,<br />
the distribution of the residuals did not appear to be normal, so the interpretation<br />
of the other tests can only be justified asymptotically. Point estimates<br />
of ρ 1 , ρ 2 , and ρ 3 were all very close to −1, suggesting a very high value of<br />
α 4π , or a strong seasonal component in inflation; we also found that lagged<br />
values of yt − 1 had a more significant effect, albeit even in this case the P-<br />
value of the test H 0 : {αy = 0} vs. H 0 : {α y > 0} was between 5% and 10%.<br />
Considering the model as a whole, anyway, we find these estimates<br />
satisfactory, although not as much as those presented by Rudebush and<br />
Svensson for the United States: we conjecture that the lower precision of<br />
the estimates for Germany depended on the smaller number of observations<br />
available.<br />
3.2. Poland<br />
Opposite to Germany, we consider Poland (and the other CEECs later)<br />
as a country small enough to be affected by the international trade, so<br />
we augmented the model with the real exchange rate depreciation with<br />
respect to Germany and with the German output gap. Albeit our dataset<br />
included 1993 too, we found that both the equations suffered from residual<br />
autocorrelation, resulting in inconsistent estimates. We interpreted this as<br />
evidence of instability, because when 1994 was taken as a starting point,<br />
the residual autocorrelation was largely removed and the other diagnostic<br />
tests were broadly compatible with a stable, correctly specified model. The<br />
estimated model was<br />
⎧<br />
ŷ ⎪⎨ t = 0.020 + 0.467y t−1 − 0.0027y t−1<br />
(0.008) (0.142) (0.001)<br />
⎪⎩<br />
̂π t =−0.64π t−1<br />
(0.12)<br />
− 0.67π t−2<br />
(0.101)<br />
− 0.51π t−3 + 21.46(q t−1 − q t−5 )<br />
(0.10)<br />
(7.88)<br />
over the period 1994Q1–2004Q1 for the AD, and 1994Q1–2004Q1 for<br />
the PC.
188 Fabrizio Iacone and Renzo Orsi<br />
The variables referred to the trade partners were then not significant in<br />
the AD equation: we found a certain effect at least for the real exchange rate<br />
depreciation, and with a sign consistent with the macroeconomic theory,<br />
but the realization of the t statistic was still rather below the critical value.<br />
The absence of the two variables referring to the international trade could<br />
be interpreted as a sign that Poland is still large enough to be more sensible<br />
to the internal rather than to the international developments, but notice that<br />
the situation was reversed for the PC equation, where only the external<br />
conditions had a significant effect on the inflationary dynamics (again we<br />
found that the effect of the omitted explanatory variable, y t−1 in this case,<br />
had the sign predicted by the macroeconomic theory, but it failed to be<br />
significantly different than 0).<br />
We concluded performing a Forecast test for a break in 2000Q2 in<br />
the PC equation, rejecting it (P-value 0.65). Since at that point, the Polish<br />
national bank abandoned the exchange rate commitment and left the direct<br />
inflation targeting to stand alone, it is fair to conclude that the change of<br />
monetary target did not result in a destabilization of the inflation dynamics.<br />
Since the target for monetary policy is defined mainly because it is assumed<br />
that in that way the expected inflation can be favorably influenced, we prefer<br />
not to give a structural interpretation of a backward looking formulation<br />
of the inflation equation, but in this case we can at least conclude that the<br />
change of target did not result in a worsening of the expectation of inflation.<br />
Thus, it is likely that by 2000, the Polish central bank acquired sufficient<br />
credibility to move away from a commitment that would have exposed it<br />
to pressure from market speculation.<br />
Although we summarized the diagnostic tests, concluding that the instability<br />
that may be associated to the transition to a market economy can be<br />
largely confined to the period before 1994, we also estimated a more restrictive<br />
specification, in which the sample for the AD equation is from 1995<br />
onwards, and for the PC from 1998 onwards. We considered this second<br />
specification, because the residuals were not normally distributed, and an<br />
asymptotic interpretation of the outcome of the autocorrelation test for the<br />
AD equation was particularly dubious because the test statistic reached<br />
approximately, the limit critical value, the P-value being slightly lower<br />
than 0.05 (it was 0.046). As far as the PC equation, the Chow test did not<br />
identify any break in 1998Q1 but the CUSUMsq statistic did indeed signal<br />
a potential break in this year, albeit only mildly. The estimates in the more
estrictive model are<br />
⎧<br />
ŷ t = 0.021<br />
⎪⎨<br />
+ 0.454y t−1<br />
(0.009)<br />
⎪⎩<br />
Inflation Control in Central and Eastern European Countries 189<br />
(0.154)<br />
̂π t =−0.57π t−1<br />
(0.19)<br />
− 0.0029y t−1<br />
(0.001)<br />
− 0.76π t−2<br />
(0.13)<br />
− 0.47π t−3 + 16.09(q t−1 − q t−5 )<br />
(0.20)<br />
(7.23)<br />
over the period 1995Q1–2004Q1 for the AD, and 1998Q1–2004Q1 for<br />
the PC, very similar then to the estimates on the longer sample. For the PC<br />
equation, we also present the alternative specification<br />
̂π t =−0.66π t−1<br />
(0.20)<br />
− 0.78π t−2<br />
(0.13)<br />
+ 19.22(q t−1 − q t−5 )<br />
(7.46)<br />
− 0.59π t−3 + 39.24π t−1<br />
(0.20) (29.00)<br />
estimated over the sample 1998Q1–2004Q1: the lagged output gap still<br />
appeared in the determinants of the inflation dynamics, it had the correct<br />
sign and it was also marginally significant with the 10% critical value and a<br />
one-sided t test: a possible interpretation of this result could be that by 1998,<br />
the PC equation evolved even closer to the standard textbook specification,<br />
but we think that a longer dataset is needed to address this particular issue.<br />
In any case, considering the whole model, we confidently conclude that<br />
the Polish monetary authority was able to control inflation during these<br />
years, but we think it mainly worked through the exchange rate management,<br />
the evidence of the closed economy transmission mechanism, in fact,<br />
being dubious.<br />
3.3. Czech Republic<br />
Czech Republic originated in 1993, from the split of the former Czechoslovakia.<br />
Data availability is then slightly reduced, and it is also possible that<br />
the division of the country acted as a source of instability additional to the<br />
one generated to the transition. Although it is not possible to distinguish<br />
between these two causes, the evidence of instability of the macroeconomic<br />
structure was quite clear.<br />
The sample for the estimation of theAD equation started in 1994Q4, but<br />
we found that a break took place around 1999, according to the CUSUMsq<br />
statistic. This is approximately also the point for which the Chow test is<br />
to be computed, and we found strong evidence in favor of it (the P-value<br />
associated to it was 0.01). Comparing the estimates in the two sub-samples,
190 Fabrizio Iacone and Renzo Orsi<br />
the most relevant differences were in the fact that the coefficient of the real<br />
rate, that signals the transmission of the monetary policy impulse from the<br />
monetary authority to the economy, and the real exchange rate depreciation,<br />
were only significant and with the correct sign in the second part of the<br />
sample (the economic activity in Germany remained insignificant).<br />
It is possible that instability affected the PC equation too, albeit in<br />
this case the evidence was less compelling. Using the whole sample that<br />
started in 1994, the CUSUMsq statistic did not indicate the presence of any<br />
break, but according to the Chow test a discontinuity took place in 1998.<br />
The estimated coefficient of the economic activity was largely insignificant,<br />
despite having the same sign predicted by the economic theory, while we<br />
found that the real exchange rate appreciation had a strongly significant<br />
anti-inflationary effect. The estimated model was<br />
⎧<br />
ŷ ⎪⎨ t = 0.010 + 0.620y t−1<br />
(0.006) (0.116)<br />
⎪⎩ ̂π t =−0.76π t−1<br />
(0.23)<br />
− 0.03y t−1 − 0.083(q t−1 − q t−5 )<br />
(0.001)<br />
(0.050)<br />
− 0.21π t−2<br />
(0.25)<br />
− 0.31π t−3 + 32.55(q t−1 − q t−5 )<br />
(0.18)<br />
(11.97)<br />
over the period 1998Q1–2004Q1 for both the AD and the PC, while if we<br />
relied on the CUSUMsq statistic rather than on the Chow one for the PC,<br />
the PC equation estimated over the period 1994Q1–2004Q1 was<br />
̂π t =−0.84π t−1<br />
(0.15)<br />
− 0.33π t−2<br />
(0.17)<br />
− 0.34π t−3 + 25.338(q t−1 − q t−5 )<br />
(0.09)<br />
(9.186)<br />
Comparing these estimates, notice that if a break did indeed take place,<br />
then the consequence was that the real exchange rate appreciation had a<br />
stronger effect against inflation in the last part of the sample, that was<br />
exactly when Czech central bank used inflation rather than exchange rate<br />
targeting. As in the case of Poland, we then found no evidence that the<br />
switch of monetary policy target resulted in a worsening of the inflationary<br />
dynamics. On the contrary, if a change in the dynamics of the growth rate<br />
of prices took place at all, when inflation targeting was introduced, it was<br />
in favor of an easier control of it.<br />
3.4. Slovenia<br />
As we did for Czech Republic, in the case of Slovenia too we only considered<br />
the part of the sample corresponding to the existence of an independent<br />
state.
Inflation Control in Central and Eastern European Countries 191<br />
The data available for the estimation started from 1994Q1 for the AD<br />
equation and in 1993Q2 for the PC. We found anyway, evidence of model<br />
mis-specification when the first year is included, resulting in residual autocorrelation<br />
and inconsistent estimates: it is also fair to suspect, on the basis<br />
of the CUSUMsq statistic, that a break took place in the first part of the<br />
sample. We then removed 1993 from the dataset, and estimated the model<br />
from 1994 onwards (sample period 1994Q1–2003Q3 for theAD, 1994Q1–<br />
2003Q4 for the PC), obtaining<br />
⎧<br />
ŷ ⎪⎨ t = 0.0226 + 0.343w t−1<br />
(0.167) (0.234)<br />
⎪⎩<br />
̂π t =−0.60π t−1<br />
(0.14)<br />
− 0.66π t−2<br />
(0.12)<br />
− 0.45π t−3 + 52.64(q t−1 − q t−5 )<br />
(0.12)<br />
(15.60)<br />
The diagnostic tests confirmed that both the equations were correctly<br />
specified and stable in the period considered, albeit, we acknowledge that<br />
the estimates in the AD equation are only significant using a 10% threshold.<br />
We are inclined to consider them anyway, suspecting that a more clear<br />
link will emerge in the future, with a longer sample and an even higher<br />
integration in the European economy. Notice, on the other hand, the absence<br />
of the real interest rate, whose estimated coefficient even failed to have the<br />
sign prescribed in the economic theory: this supported the argument of<br />
Capriolo and Lavraac that, until recently, the interference of the central<br />
bank on the financial markets made the interest rates not informative about<br />
the monetary stance.<br />
The estimated PC is qualitatively similar to the Polish and Czech one,<br />
with inflation only driven by the exchange rate dynamics but not by the<br />
output gap: indeed, ˆβ y even resulted as negative, so in the final specification<br />
we excluded y t−1 from the explanatory variables altogether. In order to see,<br />
if the more active exchange rate policy resulted in an easier inflation control<br />
and then in a larger β q , we also tested (with a Forecast test) for a break in<br />
2001, but we did not reject the hypothesis of stability.<br />
3.5. Some Concluding Remarks<br />
Poland, Czech Republic, and Slovenia are small economies that in a few<br />
years re-shaped their productive and distributive structures, opened to the<br />
international trade, established a functioning market economy and joined<br />
the EU. Despite these common features, they differ quite remarkably for<br />
the time and the condition, in which they started the transition to the market<br />
economy, for the policies they implemented since 1989, and for the size
192 Fabrizio Iacone and Renzo Orsi<br />
and the relative importance of the trade with the rest of the EU. It is then<br />
fair to expect that both the similarities and the differences emerge, and this<br />
is indeed the case.<br />
The most relevant common feature is the presence of a certain instability<br />
in the first part of the sample. To appreciate the importance of the transition,<br />
we observed that, according to a study of Coricelli and Jasbec (2004), the<br />
characteristics that may be associated to the initial conditions were the main<br />
driving forces of the real exchange rate dynamics. The variables that were<br />
more directly related to the economic notions of supply and demand, only<br />
acquired importance after a period of approximately five years. We found<br />
that the macroeconomic model structure begun to show a stable pattern<br />
only from 1994 onward, confirming the findings of Coricelli and Jasbec.<br />
Even taking 1994 as a starting value, anyway, some elements of instability<br />
remained, especially with respect to Czech Republic.<br />
The instability of the estimated macroeconomic relations is also informative<br />
about the speed of the transition. In this respect, we found that in<br />
Slovenia the adverse effect of theYugoslavian civil war canceled the advantage<br />
due to the self-management reform; Czech Republic, which had to face<br />
not only the decentralization and the break-up of CMEA but even the split<br />
of Czechoslovakia, fared worse.<br />
The relation between the variables got closer to the standard textbook<br />
description of a market supply and demand structure in the last few years.<br />
Yet, even in that part of the sample evidence of a transmission mechanism<br />
of monetary policy based on the interest rates was only present for<br />
Poland, and that too turned out to be very weak. Although it is not surprising<br />
that the exchange rate has a prominent role in inflation control in small<br />
open economies, our findings cannot be explained simply by this argument,<br />
because we explicitly include the exchange rate in both the equations of<br />
the model, then taking it into account (and holding it fixed in the interpretation<br />
of the results) in the multiple regression framework. One has rather<br />
to conclude that the interest rate channel of transmission of monetary policy<br />
was weak or obstructed, as indeed explicitly argued for Slovenia. This<br />
also means that, although these economies were acknowledged as functioning<br />
market economies, they may still be lagging behind; nonetheless,<br />
since a good development took place during the accession period, it is fair<br />
to expect that participation to the EU will foster integration. Meanwhile,<br />
their presence should not hamper inflation control too much, because the<br />
exchange rate is clearly a valid instrument for that purpose: this means that<br />
in a widened euro-zone, the Eurosystem may control inflation in the CEECs<br />
by controlling it in the core area.
Inflation Control in Central and Eastern European Countries 193<br />
Although they are all small open economies, a pattern can be envisaged<br />
in terms of dimension and openness to international trade, and this<br />
is reflected in the estimated models: Poland, is the only country in which<br />
the international factors did not affect the AD. Another way to appreciate<br />
the importance of the international openness is to look at the estimated<br />
effect of a 1% appreciation of the exchange rate: the impact on the inflation<br />
dynamics is clearly ranked according to the relative dimension of trade.<br />
Stability analysis is also important because it allows us to understand if<br />
the switch in Poland and in Czech Republic from exchange rate to inflation<br />
targeting affected the price setting formation. We did not find any evidence<br />
of it, nor we found that a more aggressive exchange rate policy in Slovenia<br />
eased the costs of disinflation from 2001 onwards. The instability that we<br />
found for Czech Republic seems to suggest an evolution of the macroeconomic<br />
structure towards a market economy, and should then have more to<br />
do with EU accession than with inflation targeting. One main implication<br />
that we perceive is that the exchange rate commitment is either irrelevant<br />
or, more likely, replaceable with inflation targeting as a nominal anchor.<br />
We think that this conclusion is of particular interest to help choosing the<br />
monetary target for the period preceding the last two years before fixing<br />
the exchange rate in the ERM2: inflation targeting should be preferred and<br />
ERM2 membership should only be limited to the two mandatory years.<br />
We also think that this lesson can be extended to other emerging market<br />
economies, and that inflation targeting, rather than explicit exchange rate<br />
commitment, should be preferred, especially if strict capital controls are<br />
not in place.<br />
References<br />
Amato, JD and S Gerlach (2002). Inflation targeting in emerging market and transition<br />
economies: lessons after a decade. European <strong>Economic</strong> Review 46, 781–790.<br />
Capriolo, G and V Lavrac (2003). Monetary and exchange rate policy in Slovenia. Eastward<br />
Enlargement of the Euro-Zone Working Papers, 17G. Berlin: Free University Berlin,<br />
Jean Monnet Centre of Excellence, 1–35.<br />
Coricelli, F and B Jasbec (2004). Real exchange rate dynamics in transition economies.<br />
Structural Change and <strong>Economic</strong> Dynamics 15, 83–100.<br />
Dvorsky, S (2000). Measuring central bank independence in selected transition countries<br />
and the disinflation process. Bank of Finland, Bofit Discussion Paper 13. Helsinki:<br />
Bank of Finland, 1–14.<br />
Iacone, F and R Orsi (2004). Exchange rate regimes and monetary policy strategies for<br />
accession countries. In Fiscal, Monetary and Exchange Rate Issues of the Eurozone<br />
Enlargement, K Zukrowska, R Orsi and V Lavrac (eds.), pp. 89–136. Warsaw: Warsaw<br />
School of <strong>Economic</strong>s.<br />
Maliszewski, WS (2000). Central bank independence in transition economies. <strong>Economic</strong>s<br />
of Transition 8, 749–789.
194 Fabrizio Iacone and Renzo Orsi<br />
Rudebush, GD and LEO Svensson (1999). Policy rules for inflation targeting. In Monetary<br />
Policy Rules, John B Taylor (ed.), pp. 203–246. Chicago: Chicago University Press.<br />
Rudebush, GD and LEO Svensson (2002). Eurosystem monetary targeting: lessons from<br />
U.S. data. European <strong>Economic</strong> Review 46, 417–442.<br />
Svensson, LEO (2000). Open-economy inflation targeting. Journal of International <strong>Economic</strong>s<br />
50, 155–183.<br />
Appendix A<br />
Summary of the Results<br />
Note to the Tables: for each equation, N indicates the normality test, AC<br />
is the LM test of no autoregressive structure in the residuals and H is the<br />
test for the heteroskedasticity; W indicates the test of those restrictions<br />
that reduce the corresponding equation as specified in general to the one<br />
specifically used. For each test, we report the P-value: the small sample<br />
version of the test is taken, unless normality is rejected. When the estimated<br />
equation is restricted, the test statistics N, AC, H, Chow, W are calculated<br />
in the unrestricted model.<br />
Table 1a.<br />
Germany.<br />
AD<br />
[sample: 91Q2–04Q1]<br />
ŷ t = 0.011 + 0.769 yt −1 − 0.038r t−1<br />
[N: 0.38] [AC: 0.20] [H: 0.17]<br />
PC<br />
[sample: 93Q1–04Q1]<br />
ˆπ t =−1.02π t−1 − 0.86π t−2 − 0.71π t−3 + 17.62y t−2<br />
[N: 0.00] [AC: 0.24] [H: 0.10] [Chow: 0.18] [W: 0.44]
Inflation Control in Central and Eastern European Countries 195<br />
Table 2a.<br />
Poland, 1994 onwards.<br />
AD<br />
[sample: 94Q1–04Q1]<br />
ŷ t = 0.020 + 0.467y t−1 − 0.0027r t−1<br />
[N: 0.00] [AC: 0.047] [H: 0.49] [Chow: 0.046] [W: 0.57]<br />
PC<br />
[sample: 94Q1–04Q1]<br />
ˆπ =−0.64π t−1 − 0.67π t−2 − 0.51π t−3 + 21.46(q t−1 − q t−5 )<br />
[N: 0.00] [AC: 0.056] [H: 0.29] [Chow: 0.36] [W: 0.06] [Chow: 2000Q2: 0.65]<br />
Table 2b.<br />
Poland, restricted sample.<br />
AD<br />
[sample: 95Q1–04Q1]<br />
ŷ t = 0.021 + 0.454y t−1 − 0.0029r t−1<br />
[N: 0.00] [AC: 0.07] [H: 0.54] [Chow: 0.54] [W: 0.64]<br />
PC<br />
[sample: 94Q1–04Q1]<br />
ˆπ t =−0.57π t−1 − 0.76π t−2 − 0.47π t−3 + 16.09(q t−1 − q t−5 )<br />
[N: 0.57] [AC: 0.25] [H: 0.79] [W: 0.39]
196 Fabrizio Iacone and Renzo Orsi<br />
Table 3.<br />
Czech Republic, restricted sample.<br />
AD<br />
[sample: 98Q1–04Q1]<br />
ŷ t = 0.010 t−1 − 0.03r t−1 + 0.083(q t−1 − q t−5 )<br />
(0.006) (0.116) (0.001)<br />
(0.050)<br />
[N: 0.79] [AC: 0.07] [H: 0.54] [W: 0.15]<br />
PC<br />
[sample: 98Q1–04Q1]<br />
ˆπ t =−0.76π t−1 − 0.21π t−2 − 0.31π t−3 + 32.55(q t−1 − q t−5 )<br />
[N: 0.71] [AC: 0.34] [H: 0.33] [W: 0.26]<br />
Table 4.<br />
Slovenia.<br />
AD<br />
[sample: 94Q1–03Q3]<br />
ŷ t = 0.226y t−1 + 0.343w t−1<br />
(0.167) (0.234)<br />
[N: 0.30] [AC: 0.37] [H: 0.86] [Chow 0.31] [W: 0.31]<br />
PC<br />
[sample: 94Q1–03Q4]<br />
ˆπ t =−0.60π t−1 − 0.66π t−2 − 0.45π t−3 + 52.64(q t−1 − q t−5 )<br />
[N: 0.44] [AC: 0.35] [H: 0.65] [Chow: 0.83] [W: 0.15] [Break 2001Q1:0.69]
Inflation Control in Central and Eastern European Countries 197<br />
Datastream codes.<br />
Czech<br />
Germany Poland Republic Slovenia<br />
Exchange rate USXRUSD<br />
(US$ per ∈)<br />
Exchange rate BDESXUSD. POI..AF. CZXRUSD. SJXRUSD.<br />
(vs. US$)<br />
CPI BDCONPRCF POCONPRCF CZCONPRCF SJCONPRCF<br />
Industrial BDINPRODG POIPTOT5H CZIPTOT.H SJIPTO01H<br />
production<br />
3-month-rate BDINTER3 PO160C.. CZ160C.. SJI60B..<br />
Diagnostic tests of some specifications that we dismissed:<br />
Table 5b.<br />
Germany, PC equation, whole dataset.<br />
PC<br />
[sample: 91Q3–04Q1]<br />
ˆπ t = 0.47 − 0.29π t−1 − 0.66π t−1 − 0.53π t−2<br />
(0.48) (0.18) (0.18) (0.17)<br />
[N:0.00] [AC: 0.20] [H: 0.06] [Chow: 0.62]<br />
− 0.50π t−3 .<br />
(0.18)<br />
+ 7.22y t−1<br />
(13.41)
198 Fabrizio Iacone and Renzo Orsi<br />
Table 5c.<br />
Poland, whole dataset.<br />
AD<br />
ŷ t = 0.02 + 0.509y t−1 − 0.0025r t−1<br />
(0.009) (0.166) (0.001)<br />
[N: 0.00] [AC: 0.20] [H:0.53] [Chow: 0.06]<br />
PC<br />
ˆπ t =−0.42 − 0.10π t−1<br />
(1.25)<br />
(0.08)<br />
− 0.130w t−1<br />
(0.203)<br />
− 0.77π t−1 − 0.78π t−2<br />
(0.12) (0.11)<br />
− 0.67π t−3 . + 22.55y t−1<br />
(0.10) (27.42)<br />
[N: 0.00] [AC: 0.01] [H: 0.07] [Chow: 0.81]<br />
+ 18.08(q t−1 − q t−5 )<br />
(9.29)<br />
[sample: 93Q2–04Q1]<br />
− 0.054(q t−1 + q t−5 )<br />
(0.045)<br />
[sample: 93Q2–04Q1]
Inflation Control in Central and Eastern European Countries 199<br />
Table 5d.<br />
Czech Republic, whole dataset.<br />
AD<br />
ŷ t = 0.007 + 0.617y t−1 − 0.0001r t−1<br />
(0.005) (0.140) (0.001)<br />
[N: 0.44] [AC: 0.94] [H: 0.06] [Chow: 0.01]<br />
PC<br />
ˆπ t = 2.45 − 0.24π t−1<br />
(1.17)<br />
(0.19)<br />
− 0.186w t−1<br />
(0.229)<br />
− 0.68π t−1 − 0.28π t−2<br />
(0.21) (0.19)<br />
+33.22(q t−1 − q t−5 )<br />
− 0.36π t−3 +22.10y t−1<br />
(0.10) (24.4)<br />
[N: 0.00] [AC: 0.25] [H: 0.086 [Chow: 0.01]<br />
[sample: 93Q4–04Q1]<br />
+ 0.047(q t−1 + q t−5 )<br />
(0.063)<br />
[sample: 94Q1–04Q1]<br />
[W:0.19]
200 Fabrizio Iacone and Renzo Orsi<br />
Table 5e.<br />
Slovenia, PC equation, whole dataset.<br />
PC<br />
ˆπ t = 4.71 − 0.50π t−1<br />
(1.41)<br />
(0.19)<br />
− 0.03π t−1 − 0.40π t−2<br />
(0.11) (0.11)<br />
[sample: 93Q2–03Q4]<br />
− 0.10π t−3 . + 2.45y t−1 + 48.28(q t−1 − q t−5 )<br />
(0.07) (26.70)<br />
(17.73)<br />
[N: 0.84] [AC: 0.03] [H: 0.02] [Chow: 0.71]
Topic 2<br />
Credit and Income: Co-Integration Dynamics<br />
of the US Economy<br />
Athanasios Athanasenas<br />
Institute of Technology and Education, Greece<br />
1. Introduction<br />
In this study, I seek to identify the existence of a significant statistical relationship<br />
between credit (as commercial bank lending) and income (GDP),<br />
for the US post-war economy, in terms of a contemporary co-integration<br />
methodology.<br />
The main empirical finding regarding the hypothesis that causality, in<br />
the long run, is directed from credit to income, agrees with the “credit-view”.<br />
This can be verified by the the post-war US data. In this chapter, I have<br />
identified a dynamic causality effect in the short run, from income changes<br />
to credit ones. To obtain the results cited here, I have applied a co-integration<br />
analysis. In addition to considering the formulated VAR, I place particular<br />
emphasis upon the stability analysis of the estimated equivalent first-order<br />
dynamic system. Finally, dynamic forecasts from the corresponding error<br />
correction variance (ECVAR) are obtained, in order to verify the validity<br />
of the co-integration vector used.<br />
This study have four sections. Section 2 reviews, in a brief note,<br />
the literature on credit and income growth of the post-war US economy.<br />
Section 3 deals with methodological issues concerning the contemporary<br />
co-integration analysis and the data used. Section 4 presents the econometric<br />
analysis and the empirical results; while Section 5 concludes the<br />
credit-income causality issue, placing emphasis on the stability analysis<br />
of the obtained first-order dynamic system and the forecasts from the<br />
corresponding ECVAR.<br />
201
202 Athanasios Athanasenas<br />
2. The Credit Issue Researched: A Brief Note<br />
on the US Economy<br />
There is an evidence that money tends to “lead” income in a historical sense<br />
(see Friedman and Schwartz (1963) and Friedman (1961; 1964). Where the<br />
mainstream monetarists through the “quantity theory” explain the empirical<br />
observations as the causal relationship running from money to income.<br />
Since the seminal works of Sims, empirical validation of the relationship<br />
between money and income fluctuations has been established (Sims, 1972;<br />
1980a). Since the publication of the works of Goldsmith (1969), Mckinnon<br />
(1973), and Shaw (1973), the debate has been going on about the role of<br />
financial intermediaries and bank credit in promoting the long-run growth.<br />
On the basis of historical data, Friedman and Schwartz (1963) argue that<br />
major depressions of the US economy have been caused by autonomous<br />
movements in money stock. 1 Sims (1980a) concludes that on one hand,<br />
money stock emerges as causally prior accounting for a substantial fraction<br />
of income variance during the inter-war and post-war periods of the business<br />
cycles. On the other hand, old skepticism remains on the direction of<br />
causality between money and income.<br />
It is well known that traditional monetarists are consistent with the<br />
old-classic view that only “money matters”, so that even when tightening<br />
of bank credits does occur during recessions, they see these as part<br />
of the endogenous financial system, rather than exogenous events, which<br />
induce recessions. As Brunner and Meltzer (1988) argue, banking crises<br />
are endogenous financial forces, which directly affect business cycles conditional<br />
upon the monetary propagation mechanism. 2<br />
In contrast, the “new-credit” viewers stress further the importance of<br />
credit restrictions, while accepting the fundamental inefficiencies of the<br />
monetary policy. 3 Bernanke and Blinder (1988), in their seminal article,<br />
conclude that credit demand is becoming relatively more stable than money<br />
1 Ibid.<br />
2 Note that monetarists seem to stress mainly, the complementarity of credit and money<br />
channels of transmission, the short-run stability of the money multiplier, and the long-run<br />
neutrality of money.<br />
3 The “new-credit” view combines the “old-credit” view of the money to spend perspective,<br />
with the money to hold perspective of the money view. The “old-credit” view focused on<br />
decisions to create and spend money, on the expansion effects of bank lending, and emphasized<br />
the issue of the non-neutrality of credit money in the long-run (see also, Trautwein,<br />
2000).
Credit and Income 203<br />
demand since 1979 and throughout the 1980s, where, money demand<br />
shocks became much more important relative to credit demand shocks in<br />
the 1980s, supported by their estimation of greater variance of money.<br />
Evidently, one of the most challenging issues of the applied monetary<br />
theory is the existence and importance of the credit channel in the monetary<br />
transmission mechanism. Bernanke (1988) emphasizes that according to the<br />
proponents of the credit view, bank credit (loans) deserve special treatment<br />
due to the qualitative difference between a borrower going to the bank for<br />
a loan and a borrower raising funds by going to the financial markets and<br />
issuing stock or floating bonds. 4<br />
Kashyap and Stein (1994) leave no doubt while supporting Bernanke’s<br />
(1983), and Bernanke’s and James’s (1991) examination of the Great<br />
Depression in the United States, that the conventional explanation for the<br />
depth and persistence of the Depression is one of the strongest pieces of evidence<br />
supporting the view that shifts in loan supply can be quite important.<br />
The supporting evidence stands strong, considering also that the decline of<br />
intensity as seen in recessions are partially due to a cut-off in bank lending<br />
(Kashyap et al., 1994). 5<br />
King and Levine (1993) stress that financial indicators, such as the<br />
importance of the banks relative to the central bank, the percentage of<br />
credit allocated to private firms and the ratio of credit issued to private<br />
firms to GDP, are strongly related with growth. For the relevant role of the<br />
monetary process and the financial sector, see, in particular, (Angeloni et al.,<br />
2003; Bernanke and Blinder, 1992; Murdock and Stiglitz, 1993; Stock and<br />
Watson, 1989; Swanson, 1998).<br />
More recently, Kashyap and Stein (2000) proved that the impact of<br />
monetary policy on lending behavior is stronger for the banks with less<br />
liquid balance sheets, where liquidity is measured by the ratio of securities<br />
4 Stiglitz (1992) stresses the fact that the distinctive nature of the mechanisms by which<br />
credit is allocated plays a central role in the allocation process. He identifies several crucial<br />
reasons that banks are likely to reduce lending activity (p. 293), as the economy goes into<br />
recession, where banks’ unwillingness and the inability to make loans obviates the effect<br />
of monetary policy. Bernanke and Blinder (1992) find that the transmission of monetary<br />
policy works through bank loans, as well as through bank deposits. Tight monetary policy<br />
reduces deposits and over time banks respond to tightened money by terminating old loans<br />
and refusing to make new ones. Reduction of bank loans for credit can depress the economy.<br />
5 Both Bernanke and Blinder (1992) and Gertler and Gilchrist (1993) using innovations in<br />
the federal funds rate as an indicator of monetary policy disturbances, find that M1 or M2<br />
declines immediately, while bank loans are slower to fall, moving contemporaneously with<br />
output.
204 Athanasios Athanasenas<br />
to assets. Their principal conclusions can be narrowed down to: (a) the existence<br />
of the credit-lending channel of monetary transmission, at least for the<br />
United States is undeniable and (b) the precise quantification and accurate<br />
measurement of the aggregate loan-supply consequences of monetary policy<br />
remain very difficult in econometrics due to the large estimation biases.<br />
See also the works of Mishkin (1995), Bernanke and Gertler (1995), and<br />
Meltzer (1995), for the bank-lending channel. Lown and Morgan (2006)<br />
stress further the substantial issue of the role of fluctuations in commercial<br />
credit standards and credit cycles being correlated with real output.<br />
All previous works provide a clear evidence that is consistent with<br />
the existence of a credit-lending channel of monetary transmission. The<br />
very heart of the credit-lending view is the proposition that the Central<br />
Banks, in general, and the Federal Reserve in the United States, in particular,<br />
can shift banks’ loan supply schedules, by conducting open-market<br />
operations. 6<br />
Consequently, given the afore-mentioned empirical research on the<br />
credit-lending channel and the established relationship between financial<br />
intermediation and economic growth in general, our main focus here is to<br />
analyze in detail the causal relationship between finance and growth by<br />
focusing on bank credit and income GNP, in the post-war US economy.<br />
Analyzing in detail here means placing particular emphasis: (a) on a contemporary<br />
co-integration approach, and (b) on the stability analysis of the<br />
observed first-order dynamic system and the dynamic forecasts from the<br />
corresponding ECVAR.<br />
3. Methodological Issues and Data<br />
In the empirical analysis, for the case of the US post-war period 1957–2007,<br />
we test for the existence of causality between the total loans and leases at<br />
commercial banks, denoted by real credit, representing the development of<br />
the banking sector, and the money income, GNP, denoted by real income,<br />
representing national economic performance. Both the variables are quarterly<br />
seasonally adjusted from 1957:3 to 2007:3, i.e., 201 observations, and<br />
are expressed in real terms, indexed by 1982–1984 as base 100, in billions of<br />
US$. Considering the natural logs, I define two new variables (W i ,Y i ) such<br />
that Y i = ln(real income i ) and W i = ln(real credit i ), for i = 1, 2,...,201.<br />
6 It is noted that the credit-lending channel also requires an imperfect price adjustment and<br />
that some borrowers cannot find perfect substitutes for bank loans.
Credit and Income 205<br />
My data is obtained on line from the Federal Reserve Bank of Saint Louis,<br />
Missouri, through their open source database. 7<br />
Researchers usually employ some measure of the stock of money over<br />
GNP, as a proxy for the size of the “financial depth” (see for instance,<br />
Goldsmith, 1969; Mckinnon, 1973). This type of financial indicator does<br />
not inform us as to whether the liabilities are those of the central banks,<br />
commercial banks, or other financial intermediaries. In this way, the creditchannel<br />
issue cannot be analyzed, and the significant role of credit cannot<br />
be isolated. Moreover, this type of financial indicator is mainly applicable<br />
to the developing countries, since the financing of the economy is to a<br />
great extent carried out by the public sector and controlled mainly by its<br />
central bank.<br />
An advanced economy such as the United States, with a fully liberalized<br />
banking and financial system, provides an excellent case for testing the<br />
modern credit view such that the commercial bank credit itself has to be<br />
modeled as a fundamental, financial, and an independent monetary variable<br />
that affects money income, and hence the economic performance, in the<br />
long run.<br />
Based on the cited references, we would accept the value of commercial<br />
bank credit to the private sector, as a measure of the ability of the banking<br />
system to provide finance-led income growth. The extent to which money<br />
income growth is associated with the credit provision from the financial<br />
sector is examined through the use of contemporary co-integration methods<br />
and system stability analysis.<br />
Initially, the order of integration of the two variables is investigated<br />
using standard unit root tests (Dickey and Fuller, 1979; Harris, 1995). The<br />
next step is the computation of co-integration vectors by applying the maximum<br />
likelihood (ML) approach considering the error-correction representation<br />
of the VAR model (Harris, 1995; Johansen and Juselius, 1990). We<br />
transformed the initial VAR to an equivalent first-order dynamic system<br />
to perform the proper stability analysis, which revealed that the system<br />
under consideration is stable. It should be re-called that stability is of vital<br />
importance for obtaining consistent forecasts, and this issue is stressed<br />
appropriately in Sections 4 (4.2 and 4.6) and 5.<br />
I proceed with the utilization of the vector error-correction modeling<br />
following Engle and Granger (1987), and testing for the short-run dynamic<br />
relationship on causality effects between money and income.<br />
7 And, we are deeply thankful for this.
206 Athanasios Athanasenas<br />
I end up by applying a dynamic forecasting technique with the estimated<br />
ECVAR. The very low value of Theil’s inequality coefficient U, provides<br />
a concrete evidence that I have formulated the suitable ECVAR obtaining,<br />
thus, robust results. Our particular emphasis on the system stability is<br />
evident and it is clarified in detail in the following section.<br />
4. Co-Integration<br />
Considering the series {Y i }, {W i } and applying the Dickey–Fuller<br />
(DA/ADF) tests (Dickey and Fuller, 1979; 1981; Harris, 1995, pp. 28–47),<br />
I found that both series are not stationary, but they are integrated of order<br />
1, that is I(1). This implies that by differentiating the initial series, we<br />
get stationary ones. In other words, the series Y i = Y i − Y i−1 and<br />
W i = W i − W i−1 are stationary, that is I(0). This can be easily verified<br />
from Fig. 1, where the non-stationary series {Y i }, {W i } are also presented.<br />
Y i<br />
W i<br />
∆Y i<br />
∆W i<br />
Figure 1. The I(1) series {Y i }, {W i } and the stationary ones {Y i }, {W i }.
Credit and Income 207<br />
Since the variables Y i , W i are used in this study, it should be re-called that<br />
the parameters of a long-run relationship represent constant elasticities.<br />
It should be re-called that if two series are not stationary but they have<br />
common trends so that there exist a linear combination of these series that<br />
is stationary, which means that they are integrated in a similar way and for<br />
this reason, they are called co-integrated, in the sense that one or the other<br />
of these variables will tend to adjust so as to restore a long-run equilibrium.<br />
Also, if there is a linear trend in the data, as in the case of Y i and W i as<br />
shown in Fig. 1, then we specify a model including a time-trend variable,<br />
allowing thus the non-stationary relationships to drift.<br />
4.1. The VAR Model and The Equivalent ECVAR<br />
We start this section by formulating and estimating an unrestricted VAR of<br />
the form:<br />
p∑<br />
x i = δ + µt i + A j x i−j + w i (1)<br />
j=1<br />
where x ∈ E n (here E denotes the Euclidean space), δ is the vector of<br />
constant terms, t i is a time-trend variable, µ is the vector of corresponding<br />
coefficients, and w is the noise vector. The matrices of coefficients A j , are<br />
defined on E n × E n . It is noted that variable t i is included for the reason<br />
mentioned above. It can be easily shown (Johansen, 1995), that (1) may be<br />
expressed in the form of an equivalent ECVAR, that is:<br />
p−1<br />
∑<br />
x i = δ + µt i + Q j x i−j + x i−1 + w i (2)<br />
j=1<br />
where Q j = ( ∑ j<br />
k=1 A k − I).<br />
After estimation, we may compute matrix from:<br />
⎛ ⎞<br />
p∑<br />
= ⎝ A j<br />
⎠ − I<br />
j=1<br />
(2a)<br />
or, directly from the corresponding ECVAR as given in Eq. (2). It is re-called<br />
that this type of VAR models like the one presented above, is recommended<br />
by Sims (1980a; b), as a way to estimate the dynamic relationships among<br />
jointly endogenous variables. It should be noted that for the case under
208 Athanasios Athanasenas<br />
consideration, vector x in Eqs. (1) and (2) is 2-dimensional, i.e.,<br />
[ ]<br />
Yi<br />
x i = and it is x ∼ I(1).<br />
W i<br />
It is known that the matrix can be decomposed such that = AC,<br />
where the elements of A are the coefficients of adjustment and the rows<br />
of matrix C are the (possible) co-integrating vectors. In some books, A is<br />
denoted by α and B by β ′ , although capital letters are used for matrices.<br />
Besides, β denotes the coefficient vector in a general linear econometric<br />
model. To avoid any confusion, we adopted the notation used here, as well<br />
as in Mouza and Paschaloudis (2007).<br />
Usually, matrices A and C are obtained by applying the ML method<br />
(Harris, 1995; Johansen and Juselius, 1990, p. 78). According to this procedure,<br />
a constant and/or trend can be included in the co-integrating vectors<br />
in the following way.<br />
Considering Eq. (2), we can augment matrix to accommodate, as an<br />
additional column, the vectors δ and µ in the following way.<br />
˜ = [.µδ]. (3)<br />
In this case, ˜ is defined on E n × E m , with m>n. For conformability,<br />
the vector x i−1 in Eq. (2) should be augmented accordingly by using the<br />
linear advance operator L (such that L k y i = y i+k ), i.e.,<br />
⎡<br />
˜x i−1 = ⎣ x ⎤<br />
i−1<br />
Lt i−1<br />
⎦ .<br />
(3a)<br />
1<br />
Hence, Eq. (2) takes the form:<br />
p−1<br />
∑<br />
x i = Q j x i−j + ˜˜x i−1 + w i . (4)<br />
j=1<br />
Note that no constants (vector δ), as well as coefficients of the time trend<br />
(vector µ) are explicitly presented in the ECVAR in Eq. (4), which specifies<br />
precisely a relevant set of n ECM.<br />
It is assumed at this point that matrices C and A refer to matrix ˜ as<br />
seen in Eq. (4). Let us consider now the kth row of matrix C, that is c ′ k. .8 If<br />
this row is assumed to be a co-integration vector, then the (disequilibrium)<br />
8 Note that dot is necessary to distinguish the kth row of C, i.e., c ′ k. , from the transposed of<br />
the kth column of this matrix that is c ′ k .
Credit and Income 209<br />
errors computed from u i = c ′ k. ˜x i must be a stationary series. Considering<br />
now the kth column of matrix A, that is a k , and given that u i−1 = c ′ k. x i−1,<br />
then the ECVAR in Eq. (4) can be written as:<br />
p−1<br />
∑<br />
x i = Q j x i−j + a k u i−1 + w i . (5)<br />
j=1<br />
This is the conventional form of an ECVAR, when matrix ˜ instead<br />
of is considered. In any case, the maximum lag in Eq. (5) is p − 1. It<br />
should be noted that this specification of an ECM is fully justified from the<br />
theoretical point of view. However, if we want to include an intercept in<br />
Eq. (5), i.e., in the short-run model, it may be assumed that the intercept in<br />
the co-integration vector is cancelled by the intercept in the short-run model,<br />
leaving only an intercept in Eq. (5). Thus, when a vector of constants is to<br />
be included in Eq. (4), which, in this case, takes the form:<br />
p−1<br />
∑<br />
x i = δ + Q j x i−j + ˜˜x i−1 + w i (6)<br />
j=1<br />
then Eq. (5), matrix ˜ and the augmented vector ˜x will become:<br />
p−1<br />
∑<br />
x i = δ + Q j x i−j + a k u i−1 + w i<br />
j=1<br />
(6a)<br />
˜ = [.µ]<br />
(6b)<br />
[ ]<br />
xi−1<br />
˜x i−1 = . (6c)<br />
Lt i−1<br />
According to Eq. (6a), the co-integrating vector c ′ k.<br />
, used to compute<br />
u i = c ′ k. ˜x i, includes only the coefficient of the time-trend variable. The<br />
inclusion of a trend in the model can be justified, if we examine the estimation<br />
results of a long-run relationship considering Y and W. In such a<br />
case, we immediately realize that a trend should be present.<br />
After adopting the VAR as seen in Eq. (1), the next step is to determine<br />
the value of p from Table 1 (Holden and Perman, p. 108).<br />
We see from Table 1 that for α ≤ 0.05, p = 4, which means that we<br />
have to estimate a VAR(4), with only one deterministic terms (i.e., trend).<br />
After estimation, we found that matrix ˜ (hat is omitted for simplicity),
210 Athanasios Athanasenas<br />
Table 1. Determination of the value of p.<br />
Lag length p<br />
LR<br />
Value of p (probability)<br />
1 — —<br />
2 132.91 0.0000<br />
3 14.988 0.0047<br />
4 14.074 0.0071<br />
5 2.7328 0.6035<br />
6 5.1465 0.2726<br />
7 2.3439 0.6728<br />
8 5.2137 0.2661<br />
specified in Eq. (6b) has the following form:<br />
[ Y i W i t i ]<br />
−0.092971 0.028736 0.000320<br />
˜ =<br />
. (7)<br />
0.016209 −0.024594 0.000124<br />
4.2. Dynamic System Stability<br />
TheVAR(4) can be transformed to an equivalent first-order dynamic system,<br />
in the following way.<br />
⎡ ⎤ ⎡<br />
⎤ ⎡ ⎤<br />
x i A 1 A 2 A 3 A 4 x i−1<br />
⎢ Lx i<br />
⎣ L 2 ⎥<br />
x i<br />
⎦ = ⎢ I 2 0 0 0<br />
⎥ ⎢ Lx i−1<br />
⎣ 0 I 2 0 0 ⎦ ⎣ L 2 ⎥<br />
x i−1<br />
⎦<br />
L 3 x i 0 0 I 2 0 L 3 x i−1<br />
⎡ ⎤ ⎡ ⎤ ⎡ ⎤<br />
δ µ w i<br />
+ ⎢ 0<br />
⎥<br />
⎣ 0 ⎦ + ⎢ 0<br />
⎥<br />
⎣ 0 ⎦ t i + ⎢ 0<br />
⎥<br />
⎣ 0 ⎦ . (8)<br />
0 0 0<br />
L in this place, denotes the linear lag operator, such that L k z i = z i−k .<br />
Equation (8) can be written in a compact form as:<br />
y i = Ãy i−1 + d + zt i + v i<br />
(8a)<br />
where<br />
y i ′ = [x i Lx i L 2 x i L 3 x i ], d ′ = [δ 0 0 0],<br />
z ′ = [µ 0 0 0], v i ′ = [w i 0 0 0]<br />
and the dimension of matrix à is (8 × 8).
Credit and Income 211<br />
Table 2.<br />
Eigen values.<br />
No. Real part Coefficient of imaginary part Length<br />
1 0.94 0 0.94<br />
2 0.8567 0.0579 0.8586<br />
3 0.8567 −0.0579 0.8586<br />
4 0.3807 0.0 0.3807<br />
5 0.0367 0.3655 0.3674<br />
6 0.0367 −0.3677 0.3674<br />
7 −0.3581 0.0 0.3581<br />
8 −0.1142 0.0 0.1142<br />
To decide whether the dynamic system described by Eq. (8a) is stable, we<br />
compute the eigenvalues of matrix Ã, presented in Table 2.<br />
Since the greater length (0.94) is less then 1, the dynamic system as<br />
given in Eq. (8a) is stable and can be used for forecasting purposes. A side<br />
verification of this stability test, is that once the system is stable, then the<br />
elements of the state vector x in the initial VAR(4), which in this case are the<br />
series {Y i }, {W i }, are either difference stationary series (DSS), or in some<br />
cases, stationary series. In fact, we show that these series are DSS and in<br />
particular, they are I(1).<br />
4.3. The Estimated Co-Integrating Vectors<br />
We applied the ML method, to compute matrices C and A, which have the<br />
following form:<br />
Y W t<br />
[ ]<br />
1 −0.289613 −0.003621<br />
C =<br />
,<br />
1 −0.653108 −0.000292<br />
[ ]<br />
−0.087991 −0.004980<br />
A =<br />
−0.038535 0.054745<br />
(9)<br />
satisfying AC = ˜. We can easily verify that only the first row of matrix<br />
C is a promising candidate, verified from the computed value of the t-<br />
statistic (−0.23), which corresponds to the (1, 2) element of matrix A, that<br />
corresponds to the second row of matrix C. Thus, we have the co-integrating
212 Athanasios Athanasenas<br />
vector:<br />
[1 − 0.289613 − 0.003621] (10)<br />
producing the errors u i , which are the disequilibrium errors obtained from:<br />
u i = Y i − 0.289613W i − 0.003621t i .<br />
(10a)<br />
It should be noted that the vector specified in Eq. (10) is a co-integration<br />
vector, iff the series {u i } computed from Eq. (10a) is stationary. To find out<br />
that this series is stationary, we run the following regression with a constant<br />
term, since ∑ u i ̸= 0.<br />
q∑<br />
u i = a + b 1 u i−1 + b j+1 u i−j + ε i . (11)<br />
Note that the value of q is set such that the noises ε i to be white. With q = 3,<br />
the estimation results are as follows:<br />
3∑<br />
u i = 0.4275 − 0.07384 u i−1 + ˆb j+1 u i−j +ˆε i<br />
(0.1174) (0.02029) (11a)<br />
j=1<br />
t =−3.64.<br />
It should be noted that in Eq. (11a), there is no problem regarding<br />
autocorrelation and heteroscedasticity. We have to compute the t u -statistic<br />
from MacKinnon (1991, pp. 267–276) critical values (see also Granger and<br />
Newbold, 1974; Hamilton, 1994; Harris, 1995, Table A6, p. 158; Harris,<br />
1995, pp. 54–55). This statistic (t u = ∞ + 1 /T + 2 /T 2 , where T<br />
denotes the sample size) is evaluated from the relevant table, taking into<br />
account Eqs. (10a) and (11a). Note that in the latter equation, there is only<br />
one deterministic term (constant). The value of the t u -statistic for this case<br />
is −3.3676 (α = 0.05). Hence, since −3.64 < −3.3676, we reject the<br />
null that the series {u i } is not stationary. Recall that the null [u i ∼ I(1)] is<br />
rejected in favor of H 1 [u i ∼ I(0)], if t
Credit and Income 213<br />
4.4. Essential Remarks<br />
In seminal works about money and income (as for instance in Tobin, 1970),<br />
the analysis is restricted in a pure theoretical consideration. Some authors<br />
(see for instance, Friedman and Kuttner, 1992) write many explanations<br />
about co-integration, co-integrating vectors, and error-correction models,<br />
but I did not trace any attempt to compute either. In a later paper by the<br />
same authors (Friedman and Kuttner, 1993), I read (p. 200) the phrase,<br />
“vector autoregression system,” which in fact is understood as a VAR. The<br />
point is that no VAR model is specified in this work. Besides, the lag length<br />
is determined by simply applying the F-test (p. 191). In this context, the<br />
validity of the results is questionable. In other applications (see for instance,<br />
Arestis and Demetriades, 1997), where a VAR model has been formulated,<br />
nothing is mentioned about the test applied to determine the maximum lag<br />
length. Also, in this paper, the authors are restricted to the use of trace<br />
statistic (l (trace) ) to test for co-integration (p. 787), although this is the<br />
necessary but not the sufficient condition, as pointed out above. From other<br />
research works (see for instance, Arestis et al., 2001), I get the impression<br />
that there is not a clear notion regarding stationarity and/or integration,<br />
since the authors state [p. 22 immediately after Eq. (1)], that “D isaset<br />
of I(0) deterministic variables such as constant, trend and dummies. . . ”.<br />
It is clear that this verification is false. How can a trend, for instance, be<br />
a stationary, I(0), variable? 10 The authors declare (p. 24) that “We then<br />
perform co-integration analysis. . . ”. The point is that apart from some<br />
theoretical issues, I did not trace such an analysis in the way it is applied<br />
here. Besides, nothing is mentioned about any ECVAR, analogous to the one<br />
that is formulated and used in this study. Most essentially, I have not traced,<br />
In case that the disequilibrium errors are computed from:<br />
then the co-integration vector will have the form:<br />
u i = Ŷ i − Y i (13)<br />
[−10.289613 0.003621]. (14)<br />
It is noted also that hat (ˆ) is omitted from the disequilibrium errors u i , for avoiding any<br />
confusion with the OLS disturbances.<br />
It should be emphasized that in such a case, i.e., using the disequilibrium errors computed<br />
from Eq. (13), then the sign of the coefficient of adjustment, as will be seen later, will be<br />
changed.<br />
10 It should be re-called at this point, that if t i is a common trend, then t i = 1(∀ i).
214 Athanasios Athanasenas<br />
in relevant works, the stability analysis applied for the system presented<br />
in Eq. (8).<br />
4.5. The ECM Formulation<br />
The lagged values of the disequilibrium errors, that is u i−1 , serve as an<br />
error correction mechanism in a short-run dynamic relationship, where the<br />
additional explanatory variables may appear in lagged first differences. All<br />
variables in this equation, also known as ECM, are stationary so that, from<br />
the econometric point of view, it is a standard single equation model, where<br />
all the classical tests are applicable. It should be noted, that the lag structure<br />
and the details of the ECM, should be in line with the formulation as seen<br />
in Eq. (6a). Hence, I started from this relation considering the errors u i<br />
and estimated the following model, given that the maximum lag length is<br />
p − 1 i.e., 3.<br />
Y i = α 0 +<br />
3∑<br />
α j Y i−j +<br />
j=1<br />
3∑<br />
β j W i−j − a Y u i−1 + Y v i . (15)<br />
j=1<br />
Note that Y v i are the model disturbances. If the adjustment coefficient<br />
a Y is significant, then we may conclude that in the long run, W causes Y.<br />
If a Y = 0, then no such a causality effect exists. In case that all β j are<br />
significant, then there is a causality effect in the short run, from W to<br />
Y. Ifallβ j = 0, then no such causality effect exists. The estimation<br />
results are presented below.<br />
Y i = 0.5132+ 0.2618Y i−1 + 0.161Y i−2 − 0.00432Y i−3 + 0.0273W i−1<br />
(0.125) (0.074) (0.073) (0.0738) (0.075)<br />
p value 0.0001 0.0005 0.0276 0.953 0.719<br />
Hansen 0.0370 0.1380 0.2380 0.234 0.128<br />
+ 0.0487W i−2 − 0.076W i−3 − 0.0879u i−1 + Y ˆv i<br />
(0.075) (0.065) (0.022)<br />
p value 0.519 0.242 0.0001<br />
Hansen 0.065 0.098 0.025 (for all coefficients 2.316) (16)<br />
¯R 2 = 0.167,s= 0.009, DWd = 1.98,F (7,189) = 6.65,<br />
Condition number (CN) = 635.34.<br />
(p value = 0.0)
Credit and Income 215<br />
We observe that the estimated adjustment coefficient (−0.0879) is<br />
significant and almost identical to the (1, 1) element of matrix A, as seen<br />
in Eq. (9), which is computed by the ML method. According to Hansen<br />
statistics, all coefficients seem to be stable for α = 0.05. The BG test<br />
revealed that no autocorrelation problems exist. In models like the one seen<br />
in Eq. (16), the term u i−1 usually produces the smallest Spearman’s correlation<br />
coefficient (r s ). In this particular case, we have: r s =−0.0979,<br />
t =−1.37, p = 0.172, which means that no heteroscedasticity problems<br />
exist. Finally, the RESET test shows that there is no any specification error.<br />
In Eq. (16), we have an inflated CN due to spurious multicollinearity,<br />
which is verified from the revised CN, having the value of 3.42 and<br />
as Lazaridis (2007) has shown, this means that, in fact, there is not any<br />
serious multicollinearity problem, affecting the reliability of the estimation<br />
results. In many applications, particularly when the variables are in logs,<br />
then usually the value of this statistic (CN) is extremely high, which in most<br />
cases, as stated above, is due to the spurious multicollinearity, which can be<br />
revealed by computing the revised CN, which is not widely known and it<br />
seems that this is the main reason, for not reporting this statistic in relevant<br />
applications.<br />
To test the null, H 0 : β 1 = β 2 = β 3 = 0, where β j are the coefficients of<br />
W i−j (j = 1, 2, 3), we computed the F-statistic that is F (3,189) = 0.533<br />
(p- value = 0.66). Hence, the null is accepted, which means that there is not<br />
any short-run causality effect from W to Y, but only in the levels, i.e., in<br />
the long run, W causes Y. Observing the value of the adjustment coefficient<br />
(−0.0879), I may conclude that it gives a satisfactory percentage, regarding<br />
the money income convergence towards a long-run equilibrium. 11<br />
4.6. Dynamic Forecasting with an ECVAR<br />
To complete the ECVAR model, it is necessary to estimate a short-run relation<br />
for W. Thus, we consider an equation which is similar to Eq. (15) i.e.,<br />
W i = b 0 +<br />
3∑<br />
a j Y i−j +<br />
j=1<br />
3∑<br />
b j W i−j − a W u i−1 + W v i . (17)<br />
j=1<br />
11 It is already mentioned that if the disequilibrium errors are computed from Eq. (13),<br />
instead of Eq. (12), then the estimated coefficient of adjustment will have a positive sign.
216 Athanasios Athanasenas<br />
The estimation results are as follows 12 :<br />
W i = 0.2241+ 0.0977Y i−1 + 0.2256Y i−2 + 0.199Y i−3 + 0.4864W i−1<br />
(0.125) (0.0733) (0.072) (0.073) (0.075)<br />
p value 0.075 0.183 0.002 0.0072 0.000<br />
Hansen 0.075 0.181 0.041 0.7130 0.045<br />
+ 0.0123W i−2 + 0.068W i−3 − 0.03852u i−1 + W ˆv i<br />
(0.87) (0.29) (0.0216)<br />
p value 0.519 0.242 0.076<br />
Hansen 0.062 0.181 0.075 (for all coefficients 2.268)<br />
¯R 2 = 0.541, s = 0.009, DWd = 1.97,<br />
F (7,189) = 34.0(p value = 0.0), CN = 635.4, Revised CN = 3.42.<br />
(18)<br />
Presumably, we are on the right way, since the estimated adjustment coefficient<br />
is almost the same as the one computed by the ML method as in<br />
the previous case. This is the (2, 1) element of matrix A as seen in Eq. (9).<br />
Note that this coefficient is significant for α ≥ 0.08, which means that there<br />
is a rather weak causality effect from Y to W in the long run.<br />
To test the null H 0 : a 1 = a 2 = a 3 = 0, where the coefficients<br />
a j (j = 1, 2, 3) as seen in Eq. (17), we compute the relevant F-statistic,<br />
i.e., F (3,189) = 8.42 (p-value = 0.0). This implies that indeed there is a<br />
causality effect in the short run, from Y to W.<br />
Considering the complete ECVAR, i.e., Eqs. (16) and (18), Eq. (10a) is<br />
used to compute the series {u i }, together with some trivial identities, I form<br />
the following system:<br />
Y i = ˆβ 0 +<br />
W i = ˆb 0 +<br />
3∑<br />
ˆα j Y i−j +<br />
j=1<br />
3∑<br />
â j Y i−j +<br />
j=1<br />
3∑<br />
ˆβ j W i−j −â Y u i−1 + Y ˆv i<br />
j=1<br />
3∑<br />
ˆb j W i−j −â W u i−1 + W ˆv i<br />
j=1<br />
12 According to Hansen statistics, all coefficients seem to be stable for α = 0.05.<br />
The Spearman’s correlation coefficient (r s ) regarding the term u i−1 , is: r s = −0.075,<br />
t =−1.051, p = 0.294, which means that we do not have to bother about heteroscedasticity<br />
problems. Also the revised CN indicates that no multicollinearity problems exist, in<br />
order to affect the reliability of the estimation results.
Credit and Income 217<br />
Y i = Y i + Y i−1<br />
W i = W i + W i−1<br />
u i = Y i − 0.289613W i − 0.0036218t i .<br />
(19)<br />
It may be useful to note that there are three identities in the system<br />
and the only exogenous variable present in the system is the trend. We may<br />
obtain dynamic simulation results from the ECVAR specified in Eq. (19),<br />
in order to obtain predictions for Y i and W i and at the same time to verify<br />
the validity of the co-integration vector used. This can be achieved by first<br />
formulating the following deterministic system.<br />
K 0 y i = K 1 y i−1 + K 2 y i−2 + K 3 y i−3 + ˜Dq i (20)<br />
where y i = [ϒ i W i u i Y i W i ] ′ , q i = [t i 1] ′<br />
⎡<br />
K 0 =<br />
⎢<br />
⎣<br />
1 0 0 0 0<br />
0 1 0 0 0<br />
0 0 1 −1 0.289613<br />
−1 0 0 1 0<br />
0 −1 0 0 1<br />
⎤<br />
⎥<br />
⎦ ,<br />
⎡<br />
⎤<br />
0.2618 0.0273 −0.0879 0 0<br />
0.0977 0.4864 −0.03852 0 0<br />
K 1 =<br />
⎢ 0 0 0 0 0<br />
⎥<br />
⎣ 0 0 0 1 0⎦ ,<br />
0 0 0 0 1<br />
⎡<br />
⎤<br />
0.161 0.0487 0 0 0<br />
0.2256 0.0123 0 0 0<br />
K 2 =<br />
⎢ 0 0 0 0 0<br />
⎥<br />
⎣ 0 0 0 0 0⎦ ,<br />
0 0 0 0 0<br />
⎡<br />
⎤<br />
−0.00432 0.076 0 0 0<br />
0.199 −0.068 0 0 0<br />
K 3 =<br />
⎢ 0 0 0 0 0<br />
⎥<br />
⎣ 0 0 0 0 0⎦ ,<br />
0 0 0 0 0
218 Athanasios Athanasenas<br />
and<br />
⎡<br />
˜D =<br />
⎢<br />
⎣<br />
⎤<br />
0 0.5132<br />
0 0.2241<br />
⎥<br />
⎦ .<br />
−0.003621 0<br />
0 0<br />
0 0<br />
Pre-multiplying Eq. (20) by K0 −1 we get:<br />
y i = Q 1 y i−1 + Q 2 y i−2 + Q 3 y i−3 + Dq i (21)<br />
where Q 1 = K0 −1 K 1, Q 2 = K0 −1 K 2, Q 3 = K0 −1 K 3, and D = K0<br />
−1 ˜D.<br />
The system given in Eq. (21) can be transformed to an equivalent firstorder<br />
dynamic system, in a similar way to the one already described earlier<br />
[see Eqs. (8) and (8a)]. It is noted that in this case matrix à is of dimension<br />
(15 × 15). It is important to mention again that we should avoid using<br />
a dynamic system, for simulation purposes, that is unstable. Thus, from<br />
this system, we can obtain dynamic simulation results for the variables Y i<br />
and W i . These results are graphically presented (Fig. 2).<br />
The very low value of Theil’s inequality coefficient U, for both cases<br />
is a pronounced evidence that, apart from computing the indicated cointegration<br />
vector, we have also formulated the suitable ECVAR so that the<br />
results obtained are undoubtfully robust.<br />
5. Conclusions and Implications<br />
The purpose of this study is to contribute to the empirical investigation of<br />
the co-integration dynamics of the credit-income nexus, within the economic<br />
growth process of the post-war US economy, over the period from<br />
1957 to 2007.<br />
Utilizing advanced and contemporary co-integration analysis and<br />
applying vector ECM estimation, we place special emphasis on forecasting<br />
and system stability analysis. I can say that to the best of my knowledge,<br />
similar system stability and forecasting analysis, as the one applied here,<br />
is very difficult to meet in the relevant literature, after taking into consideration<br />
similar research works. See for example, (Arestis and Demetriades,<br />
1997; Arestis et al., 2001; Demetriades and Hussein, 1996; Friedman and<br />
Kuttner, 1992; 1993; Levine and Zervos, 1998; Rousseau and Wachtel,<br />
1998; 2000).<br />
My results state clearly that there is no short-run causality effect from<br />
credit changes to income changes, but only in the levels, that is in the long
Credit and Income 219<br />
Figure 2.<br />
Results of dynamic simulations.
220 Athanasios Athanasenas<br />
run, credit causes money income. From my co-integration analysis, I found<br />
that the adjustment coefficient is significant and its value (−0.0879), gives a<br />
satisfactory percentage, regarding the money income convergence towards<br />
a long-run equilibrium.<br />
Moreover, we observe a causality effect from income changes to credit<br />
changes, in the short run for the post-war US economy. The validity of<br />
the credit view theorists seems rather evident, by taking into account the<br />
contemporary co-integration and system stability approaches.<br />
Further, in terms of reasonable research implications, following our<br />
contemporary co-integration and system stability approaches, we may<br />
stress the need for an in-depth investigation of the credit cycle and fluctuations<br />
in commercial credit standards. 13<br />
References<br />
Angeloni, I, KA Kashyap and B Mojon (2003). Monetary Policy Transmission in the Euro<br />
Area. Part 4, Chap. 24 NY: Cambridge University Press.<br />
Arestis, P and P Demetriades (1997). Financial development and economic growth: assessing<br />
the evidence. The <strong>Economic</strong> Journal 107, 783–799.<br />
Arestis, P, PO Demetriades and KB Luintel (2001). Financial development and economic<br />
growth: the results of stock markets. Journal of Money, Credit and Banking 33(1),<br />
16–41.<br />
Bernanke, SB (1983). Non-monetary effects of the financial crisis in the propagation of the<br />
great depression. American <strong>Economic</strong> Review 73(3), 257–276.<br />
Bernanke, SB (1988). Monetary policy transmission: through money or credit? Business<br />
Review, November/December, 3–11 (Federal Reserve Bank of Philadelphia).<br />
Bernanke, SB and SA Blinder (1988). Credit, money, and aggregate demand. The American<br />
<strong>Economic</strong> Review, Papers and Proceedings 78(2), 435–439.<br />
Bernanke, SB and SA Blinder (1992). The federal funds rate and the channels of monetary<br />
transmission. The American <strong>Economic</strong> Review 82(4), 901–921.<br />
Bernanke, SB and M Gertler (1995). Inside the black box: the credit channel of monetary<br />
transmission. The Journal of <strong>Economic</strong> Perspectives 9(4), 27–48.<br />
Bernanke, SB and H James (1991). The gold standard, deflation, and financial crisis in the<br />
great depression: an International comparison. In Financial Markets and Financial<br />
Crises, RG Hubbard (ed.), Chicago: University of Chicago Press for NBER, 33–68.<br />
Brunner, K and HA Meltzer (1988). Money and credit in the monetary transmission process.<br />
American <strong>Economic</strong> Review 78, 446–451.<br />
Demetriades, PO and K Hussein (1996). Does financial development cause economic<br />
growth? Journal of Development <strong>Economic</strong>s 51, 387–411.<br />
13 See, for example, Lown and Morgan (2006), for parallel research on commercial credit<br />
standards, following “traditional” VAR analysis.
Credit and Income 221<br />
Dickey, DA and WA Fuller (1979). Distribution of the estimators for autoregressive<br />
time series with a unit root. Journal of the American Statistical Association 74,<br />
427–431.<br />
Dickey, DA and WA Fuller (1981). Likelihood ratio statistics for autoregressive time series<br />
with a unit root. Econometrica 49, 1057–1072.<br />
Engle, RF and CWJ Granger (1987). Cointegration and error-correction: representation,<br />
estimation, and testing. Econometrica 55, 251–276.<br />
Friedman, M (1961). The lag in the effect of monetary policy. Journal of Political Economy,<br />
October 69, 447–466.<br />
Friedman, M (1964). The monetary studies of the national bureau. National Bureau of <strong>Economic</strong><br />
Research, Annual Report. New York: National Bureau of <strong>Economic</strong> Research,<br />
1–234.<br />
Friedman, B and KN Kuttner (1992). Money, income, prices and interest rates. The American<br />
<strong>Economic</strong> Review 82(3), 472–492.<br />
Friedman, B and KN Kuttner (1993). Another look at the evidence of money-income causality.<br />
Journal of Econometrics 57, 189–203.<br />
Friedman, M and A Schwartz (1963). Money and business cycles. Review of <strong>Economic</strong>s and<br />
Statistics, February, (suppl. 45), 32–64.<br />
Gertler, M and S Gilchrist (1993). The role of credit market imperfections in the monetary<br />
transmission mechanism: arguments and evidence. Scandinavian Journal of <strong>Economic</strong>s<br />
95(1), 43–64.<br />
Goldsmith, RW (1969). Financial structure and development. New Haven, CT: Yale<br />
University Press.<br />
Granger, CWJ and P Newbold (1974). Spurious regressions in econometric model specification.<br />
Journal of Econometrics 2, 111–120.<br />
Hamilton, JD (1994). Time Series Analysis. Princeton: Princeton University Press.<br />
Harris, R (1995). Using Cointegration Analysis in Econometric Modeling. London: Prentice<br />
Hall.<br />
Holden, D and R Perman (1994). Unit roots and cointegration for the economists. In Cointegration<br />
for the Applied Economist, BB Rao (Ed.), UK: St. Martin’s Press, 47–112.<br />
Johansen, S (1995). Likelihood-Based Inference in Cointegrating Vector Autoregressive<br />
<strong>Models</strong>. New York: Oxford University Press.<br />
Johansen, S and K Juselius (1990). Maximum likelihood estimation and inference on cointegration<br />
with application to the demand for money. Oxford Bulletin of <strong>Economic</strong>s and<br />
Statistics 52, 169–210.<br />
Kashyap, KA and CJ Stein (1994). Monetary policy and bank lending. NBER Studies in<br />
Business Cycles 29, 221–256.<br />
Kashyap, KA and CJ Stein (2000). What do a million observations on banks say about the<br />
transmission of monetary policy? The American <strong>Economic</strong> Review 90(3), 407–428.<br />
Kashyap, KA, AO Lamont and CJ Stein (1994). Credit conditions and the cyclical behavior<br />
of inventories. The Quarterly Journal of <strong>Economic</strong>s 109(3), 565–592.<br />
King, GR and R Levine (1993). Finance, entrepreneurship and growth: theory and evidence.<br />
Journal of Monetary <strong>Economic</strong>s 32, 1–30.<br />
Lazaridis, A (2007). A note regarding the condition number: the case of spurious and latent<br />
multicollinearity. Quality & Quantity 41(1), 123–135.<br />
Levine, R and S Zervos (1998). Stock markets, banks, and economic growth. American<br />
<strong>Economic</strong> Review 88, 537–558.
222 Athanasios Athanasenas<br />
Lown, C and DP Morgan (2006). The credit cycle and the business cycle: new findings<br />
using the loan officer opinion survey. Journal of Money, Credit, and Banking 38(6),<br />
1575–1597.<br />
MacKinnon, J (1991). Critical values for co-integration tests. In Long-Run <strong>Economic</strong> Relationships,<br />
RF Engle and CWJ Granger (eds.), pp. 267–276. Oxford: Oxford University<br />
Press.<br />
Mckinnon, RL (1973). Money and Capital in <strong>Economic</strong> Development. Washington, DC:<br />
Brookings Institution.<br />
Meltzer, HA (1995). Monetary, credit and (other) transmission processes: a monetarist perspective.<br />
The Journal of <strong>Economic</strong> Perspectives 9(4), 49–72.<br />
Mishkin, SF (1995). Symposium on the monetary transmission mechanism. The Journal of<br />
<strong>Economic</strong> Perspectives 9(4), 3–10.<br />
Mouza, AM and D Paschaloudis (2007). Advertisement and sales: a contemporary cointegration<br />
analysis. Journal of Academy of Business and <strong>Economic</strong>s 7(2), 83–95.<br />
Murdock, K and J Stiglitz (1997). Financial restraint: towards a new paradigm. Journal of<br />
Monetary <strong>Economic</strong>s 35, 275–301.<br />
Rousseau, PL and P Wachtel (1998). Financial intermediation and economic performance:<br />
historical evidence from five industrialized countries. Journal of Money, Credit and<br />
Banking 30, 657–678.<br />
Rousseau, PL and P Wachtel (2000). Equity markets and growth: cross-country evidence on<br />
timing and outcomes, 1980–1995. Journal of Banking and Finance 24, 1933–1957.<br />
Shaw, ES (1973). Financial Deepening in <strong>Economic</strong> Development. New York: Oxford<br />
University Press.<br />
Sims, CA (1972). Money, income, and causality. The American <strong>Economic</strong> Review 62(4),<br />
540–552.<br />
Sims, CA (1980a). Comparison of interwar and postwar business cycles. The American<br />
<strong>Economic</strong> Review 70, 250–277.<br />
Sims, CA (1980b). Macroeconomics and reality. Econometrica 48, 1–48.<br />
Stiglitz, EJ (1992). Capital markets and economic fluctuations in capitalist economies.<br />
European <strong>Economic</strong> Review 36, 269–306.<br />
Stock, JM and MW Watson (1988). Testing for common trends. Journal of the American<br />
Statistical Association 83, 1097–1107.<br />
Stock, J and MW Watson (1989). Interpreting the evidence on money-output causality.<br />
Journal of Econometrics, January, 161–181.<br />
Swanson, N (1998). Money and output viewed through rolling window. Journal of Monetary<br />
<strong>Economic</strong>s 41, 455–473.<br />
Tobin, J (1970). Money and income: post hoc ergo propter hoc? Quarterly Journal of<br />
<strong>Economic</strong>s 84, 301–329.<br />
Trautwein, M (2000). The credit view, old and new. Journal of <strong>Economic</strong> Surveys 14(2)<br />
155–189.
INDEX<br />
adaptive control, 44, 49, 57, 59<br />
asymptotic co-variance, 46<br />
autocorrelation, 35, 36, 40, 182, 184, 185,<br />
187, 188, 191, 212, 215<br />
automatic stabilizer, 70, 71, 74, 76, 79, 80,<br />
96, 98<br />
balance of payments, 44, 49, 53<br />
bank credit, 202–205<br />
Bank of England, 69, 70, 74, 76–78, 90<br />
British fiscal policy, 69, 73, 81<br />
budget stability, 74<br />
business cycle, 9, 202<br />
cash-balance, 52<br />
causality, 201, 202, 204, 205, 214–216,<br />
218, 220<br />
Central Bank objective function, 85<br />
Cholesky’s factorization, 24<br />
Chow stability, 184<br />
co-integration, 23, 25, 28–32, 34, 41, 201,<br />
204–206, 208, 209, 212, 213, 217, 218,<br />
220<br />
collectivism, 143<br />
conformability, 26, 208<br />
conformity, 140, 143<br />
corporate culture, 139, 145<br />
credit and income, 201, 204<br />
credit demand, 202, 203<br />
critical value, 30, 33, 38–40, 188, 189, 212<br />
current deficits, 79, 83<br />
CUSUM, 185, 186, 188–191<br />
cyclical stability, 74<br />
Czech Republic, 173–177, 181–183,<br />
189–193, 196, 197, 199<br />
debt ratio, 72, 78, 92, 96, 99<br />
decision orientation, 132<br />
decomposition, 26, 27<br />
dependent central bank, 88, 90, 94<br />
depression, 9, 202, 203<br />
deterministic, 25, 41, 209, 212, 213, 217<br />
devaluation, 57, 176, 177<br />
diagnostic, 185–188, 191, 197<br />
Dickey–Fuller tests, 30, 206<br />
discount rate, 52, 57–59<br />
discretionary fiscal policy, 82<br />
discretionary tax revenues, 82<br />
disequilibrium, 25, 28, 208, 212–215<br />
domestic absorption, 49, 50, 53, 59<br />
dynamic response multiplier, 57<br />
Eastern European Countries, 173<br />
econometrics, 6, 9, 204<br />
ECVAR, 23, 25, 201, 204, 206–209, 213,<br />
215–218<br />
Eigen values, 24–26, 211<br />
emissions, 101, 103–106<br />
enterprise integration, 121, 125–128<br />
enterprise integration modeling, 121<br />
enterprise modeling, 121–124, 132<br />
environmental economics, 101–103<br />
environmental taxation, 101, 103<br />
ERM, 176, 193<br />
error co-variance matrix, 48<br />
error-correction, 23, 201, 205, 213, 214<br />
European Commission, 69, 79, 98<br />
Eurozone, 69, 72, 77–81, 88, 93–95<br />
ex-ante responses, 76<br />
exchange rate, 50, 53, 58, 60, 74,<br />
173–178, 180–183, 185, 187–193, 197<br />
exclusivity, 143<br />
feasibility region, 3, 5<br />
feasible set, 16<br />
feedback rule, 77<br />
223
224 Index<br />
fiscal expansions, 91, 92<br />
fiscal leadership, 69, 71, 73, 76–78, 81,<br />
85, 89–97, 99<br />
fixed parity, 176<br />
function oriented enterprise, 128<br />
functional entities, 122–124, 126<br />
fund-bank adjustment policy model, 49<br />
Gaussian, 44, 62<br />
Germany, 93, 94, 99, 174, 176, 181–187,<br />
190, 194, 197<br />
goal programming, 151, 155, 156, 161,<br />
168<br />
Hamiltonian, 111<br />
health care unit, 168<br />
health service management, 151<br />
heteroscedasticity, 35, 37, 39, 184, 185,<br />
194, 212, 215, 216<br />
Hou-Ren-Sou, 140, 143, 144<br />
human resources practices, 130, 142, 144<br />
human ware, 141<br />
hybrid modeling constructs, 130<br />
imitation-type dynamics, 103, 106<br />
independent monetary policies, 71, 75<br />
Indian economy, 49, 50<br />
inflation, 69, 70, 72–74, 76–80, 82, 84, 85,<br />
88–96, 99, 153, 164, 173–179,<br />
181–183, 185–193<br />
inflation control, 70, 73, 79, 92, 173, 174,<br />
176, 177, 191, 192<br />
input-output, 6–9, 11, 12, 14<br />
institutional parameters, 76, 86, 96<br />
integration challenges, 121<br />
inter-temporal optimization problem, 107<br />
interactive voice response, 153<br />
interest rate, 38, 50–52, 57–60, 75, 77–81,<br />
176, 178, 179, 181, 191, 192<br />
Japanese national culture, 137, 139, 140,<br />
147<br />
Japanese Organizational culture and<br />
corporate performance, 137<br />
Jaque-Bera criterion, 29<br />
just in time, 144<br />
Kaizen, 140–142<br />
Kroneker product, 46<br />
leadership, 69–73, 76–79, 81, 85, 86,<br />
89–97, 99, 103, 130, 137, 139, 142,<br />
144, 145<br />
lean production system, 141, 142<br />
linear constraints, 3, 10<br />
linear programming, 3, 4, 6–8, 10, 14<br />
liquidity, 203<br />
LISREL, 145<br />
Maastricht criteria, 173, 176, 177<br />
macro values, 138, 143<br />
management support, 131<br />
mathematical programming, 3, 156<br />
maximization, 7, 82, 112, 151<br />
maximum likelihood, 24, 49, 205<br />
meso values, 138<br />
meta-analysis, 117<br />
micro values, 138, 143<br />
minimum variance, 25, 31–33, 41<br />
ML method, 24, 25, 28, 31, 32, 34, 40,<br />
208, 211, 215, 216<br />
modeling method, 124<br />
monetary and fiscal policy, 43, 44, 49, 53,<br />
56–59, 82, 84, 93<br />
monetary policy, 43, 44, 49, 53, 56–59,<br />
69–82, 84, 85, 91–95, 98, 102, 103,<br />
173, 174, 176, 178, 179, 185, 186, 188,<br />
190, 192, 202–204<br />
multicollinearity, 215, 216<br />
multi functional team, 142<br />
multiple equilibria, 101, 112, 113<br />
Nash equilibrium, 81<br />
Nash solution, 70, 102, 107, 110<br />
New Cambridge model, 50<br />
new constraint, 161, 162<br />
new credit, 202<br />
non-believers, 103, 105–107, 117<br />
object request brokers, 126<br />
old credit, 202<br />
OLS, 23, 25, 28–32, 183, 184, 213<br />
optimal control, 45, 46, 50, 55
Index 225<br />
optimal currency area, 175<br />
optimization, 3, 4, 7, 16, 44, 49, 91, 107,<br />
110–112<br />
organizational culture, 130, 137–139, 142,<br />
144<br />
output gap, 72, 77–82, 179–183, 187, 189,<br />
191<br />
Pareto-improvement, 110<br />
Pareto-inferior, 116<br />
Phillips curve, 87, 98, 178<br />
Poland, 173–177, 181, 182, 185, 187, 188,<br />
190–193, 195, 197, 198<br />
primary deficit, 72, 79<br />
pro-cyclicality, 85, 87<br />
product modeling, 122<br />
productivity, 175<br />
pseudo-inverse, 26<br />
public sector deficit, 84<br />
quadratic function, 10<br />
quality circle, 141<br />
rational expectations, 44, 86, 179<br />
reduced form, 44, 46, 48, 58, 59, 82, 84,<br />
179<br />
reduced form coefficients, 44, 46, 48<br />
regulator, 76, 101–111, 113–117<br />
regulator’s trustworthiness, 115<br />
residuals, 24, 30, 35, 36, 39, 54, 179, 184,<br />
185, 187, 188, 191, 194<br />
resource information, 122<br />
Ricatti-equations, 45<br />
robustness, 55<br />
sequential Nash game, 103, 107<br />
short-run stabilization, 70, 76, 80<br />
simplex algorithm, 6, 15<br />
singular value, 26–28, 31, 33<br />
Skiba point, 112, 115, 116<br />
Slovenia, 173–177, 181–183, 185,<br />
190–193, 196, 197, 200<br />
stability, 11, 54, 55, 57, 74, 76, 80–84, 95,<br />
113, 144, 176, 178, 184, 185, 191, 193,<br />
201, 202, 204–206, 210, 211, 214, 218,<br />
220<br />
stabilization policy, 43, 71<br />
standard deviation, 30, 32–34<br />
state variable form, 49, 50<br />
statically optimal, 109, 113<br />
steady-state, 82, 83<br />
structural deficit, 79, 80<br />
structural model, 44, 49, 178<br />
symmetric stabilization, 75<br />
tax revenue, 50, 51, 57, 58, 60, 82, 83, 95,<br />
96, 104, 106<br />
Theil’s inequality, 206, 218<br />
time varying responses, 43<br />
time-invariant econometric model, 47<br />
time-varying model, 44, 47<br />
TQM, 141, 142, 144<br />
transition, 45, 50, 54, 173–176, 178–180,<br />
185, 186, 188, 189, 191, 192<br />
transmission of monetary policy, 173, 178,<br />
179, 190, 192, 203<br />
UK government, 69<br />
underutilization, 158, 160<br />
unstable middle equilibrium, 115<br />
updating method, 44, 46<br />
US economy, 201, 202, 204, 218, 220<br />
VAR, 23, 25, 27, 28, 32, 44, 178, 201, 205,<br />
207, 213, 220